Concepts

Overviews that compare the methods for one problem across papers and help you choose between them.

718 pages

Concept3D scene graphA 3D scene graph stores a scene as nodes at several levels of meaning, joined by spatial or temporal relations. Two routes are covered: a robot pipeline that builds the layers from stereo or RGB-D and IMU data, with the upper layers parsed after the run (Kimera), and a single-image pipeline that lifts a photo to 3D boxes and turns them into spatial QA for a vision-language model (SpatialRGPT).ConceptA* searchBest-first graph search that orders states by cost-so-far plus a heuristic. Quadrotor motion-primitive planning with admissible bounds: one author-reported query expanded 376 nodes with the LQMT bound vs 2707 for Dijkstra. Other papers use A* for car-like hybrid search and kinodynamic guidance, with no optimality claim.ConceptAI auditingIndependent, evidence-based checks of AI systems against stated criteria. For LLMs, one position paper proposes three linked audit layers: the provider's governance, the model before release, and each product built on it. It gives no thresholds and no experiments, so the layers are a checklist of what to examine, not a tested procedure.ConceptAI-generated text detectionDeciding whether a text was written by a language model. Covers zero-shot likelihood-curvature scoring (DetectGPT), trained classifiers and how they transfer across domains, model-side watermarks and their extension to diffusion LMs, and how paraphrasing and spoofing break detectors. Accuracy figures are author-reported and depend heavily on whether an adversary edits the text.ConceptAction chunkingRobot policies predict and execute a block of future actions. Real-time chunking keeps consecutive blocks continuous under inference delay, and training-time variants move its cost into training. Overlapping predictions can be fused. Multi-horizon training with a learned gate picks the chunk length. Chunk overlap also flags failures, and consensus among sampled chunks picks the retry chunk.ConceptAction tokenizationTurning continuous values such as robot actions or motion clips into discrete tokens that a sequence model can predict. Covered: numbers written into a frozen LLM's prompt, per-timestep binning, a DCT-plus-BPE compressor (FAST), and a learned VQ-VAE motion tokenizer for humanoid text-to-motion (UH-1).ConceptActivation patchingCausal test that copies internal activations from a clean run into a corrupted run to find which model components carry a behaviour. Path patching finds an attention-head circuit in GPT-2 small; single-site residual patching localises a verbal-confidence signal in Gemma 3 27B. Both need white-box access and several forward passes per test.ConceptActivation sparsitySkipping the neurons that stay near zero for a given input, so that only the active ones are computed or moved. The core idea is to find a stable set of frequently active "hot" neurons and to predict the active set per token. The trade-off: the gain depends on the activation function, and it needs an offline profile and a predictor per layer.ConceptActivation steeringChanges a language model's behaviour at inference by editing its internal activations, with no retraining. Inference-Time Intervention shifts attention heads toward a truthful direction, raising TruthfulQA scores on two chat models. A fixed boost to hand-picked feed-forward values cuts GPT-2 toxicity. A sparse-autoencoder concept direction steers synthetic biography QA. All need white-box access.ConceptActive learningPick which unlabeled examples to label next, so a model reaches its target accuracy with fewer labels. Deep networks make this harder: softmax confidence is unreliable and they need many labels per round. Variants: batch uncertainty or hybrid scores, Bayesian scores (BALD, MC-dropout), core-set coverage, pseudo-label expansion, stopping rules. Trade-off: fewer labels against extra compute.ConceptActive-set methodSolve a convex QP by guessing its tight inequality constraints, solving the equality-constrained system and correcting the guess. Warm-starting from the previous step's active set needs one linear solve in most steps of one walking controller. Dense active-set solvers such as qpOASES and qpmad suit whole-body control; benchmark claims are relayed.ConceptActor-criticTwo learned parts: a critic that scores states or actions, and an actor trained on those scores. Twenty papers show the design space, from imagined rollouts and token-level critics to agent-step critics, model-based value expansion, offline TD3+BC, per-agent PPO, edge-network DDPG, online MPC policies and Bayesian GP critics. Pick by data, compute and action structure.ConceptAdaBoostCombine many weak classifiers into one vote by reweighting the training examples each round, so later rounds focus on earlier mistakes. Multi-class variant with a weak learner that needs no training. Trade-off: each round scores every candidate feature, so cost grows with pool size. Compiled from one card; the other variants are not yet covered.ConceptAdaptive haltingDeciding, per input, how much computation to run before stopping. Three sources: stop a looped model when its output settles (KL and entropy thresholds, no retraining), train a halting head on answer correctness, or skip remaining layers once dominant sub-updates agree. All are author-reported on small or synthetic tasks. None compares them with fixed depth on a common benchmark.ConceptAdaptive instance normalizationAdaIN re-scales each channel to the mean and standard deviation of a reference, with no learned parameters. Three cards use it or its scale-and-shift form: ArtFlow applies it to reversible latent features, at about 6.4 times the author-reported time of plain AdaIN; StarGAN v2 injects a per-domain style code; Drive-OccWorld conditions a layer-normalised memory. No card covers the original paper.ConceptAdaptive layer normalization (AdaLN)AdaLN makes a layer norm's scale and shift depend on a conditioning vector, so a transformer can be steered without a cross-attention path. One card uses it in a diffusion policy (ScaleDP), which regresses the scale and shift from the diffusion step and observation and drops the causal mask. Reported gains come from both changes together; no ablation isolates AdaLN.ConceptAdaptive loss weightingRe-weighting loss terms, points or noise levels so that hard or slow parts of training count for more. Covers per-point residual attention (no extra network), task weights from gradient norms or loss descent, and a fixed noise-level weight for diffusion (P2). Five cards; the numbers are author-reported, and paper and code disagree on several values.ConceptAdaptive roundingPost-training weight quantization where each weight is rounded up or down to reduce a layer's output error, instead of rounding to the nearest grid value. Needs a small unlabelled calibration set and no retraining. Author-reported: 4-bit Resnet18 weights recover most of the accuracy that nearest rounding loses.ConceptAdditive kernelA Gaussian-process kernel that sums products of one-dimensional kernels over all interaction orders, with one learned variance per order. Newton–Girard identities make the full sum O(D²). It captures interactions between input coordinates without the locality of squared-exponential kernels. Trade-off: the order sum grows with D, so it is often truncated.ConceptAdversarial evaluationScore model outputs with a separate classifier or LLM judge, or measure how far bounded input perturbations push a model's output. Variants: AdverSuc for dialogue, judge-scored attack success for multimodal models, a white-box perturbation test, model-in-the-loop filtering and rounds of adversarial data collection. Pair scores with reliability checks; human-agreement evidence is weak or missing.ConceptAdversarial robustnessAdversarial robustness is a model's ability to keep its intended behaviour under deliberate input perturbation, poisoned data or attacks on its surroundings. Twenty-one cards cover surveys, text, image, multimodal, prompt, federated, backdoor and poisoning work. Numbers are author-reported under different threat models; do not compare them.ConceptAdversarial text generationProducing text that fools a learned judge or breaks a safety behaviour. Three families: training a generator against a discriminator (policy gradient handles discrete tokens), changing an input to fool a fixed detector or classifier (paraphrasing, word substitution), and optimising prompts against a model (gradient-based tokens or an attacker LLM). Ten sources.ConceptAdversarial trainingAdversarial training fits a model on inputs perturbed by an inner attack, usually PGD. The key choice is how hard that attack is solved and when. One card compares standard PGD, a step-count curriculum and a schedule driven by a stationarity score (FOSC). Author-reported CIFAR-10 PGD-10 accuracy is 48.06% for the dynamic schedule against 44.39% for standard PGD.ConceptAgent memoryHow an LLM agent keeps and recalls information across its steps. Memory is a write, manage and read loop over three sources (current trial, past trials, external knowledge), stored as text or as weights. Shared memory can stage writes and commit only checked beliefs before irreversible actions. Results are author-reported; no open benchmark targets agent memory alone.ConceptAgent-based simulationAgent-based simulation models a population as individual agents that move, meet and change state each step, so outcomes emerge from interaction. Cards cover an unvalidated hourly SEIR model, LLM participants that never interact, LLM agents on a graph, LLM agents that farm, rob and trade, and LLM reviewers with one trait varied at a time.ConceptAir-to-ground path lossModels for the signal loss between an aerial platform and a ground user, split into a clear-line-of-sight and a blocked-path case weighted by an elevation-angle probability. Five variants (low altitude, indoor users, high altitude, cellular-to-UAV, UAV-to-UAV). The constants are not given in the surveyed source, so each model needs values from its original study or from measurement.ConceptAlgorithm selectionChoosing which search policy or starting pipeline to run on a new dataset, using results from earlier datasets. Auto-Sklearn 2.0 picks one of eight policies per dataset with a pairwise random-forest selector on two meta-features, plus a fallback. Its gain over the single best policy is small and changes sign across settings; the selector clearly beats random choice. One paper so far.ConceptAlgorithmic fairnessMakes a classifier's outcomes comparable across groups defined by a sensitive attribute. Start by choosing a group criterion, since no two common criteria can all hold together. Then pick where to intervene: repair or add data, constrain or reweight training, or adjust predictions after training. Each intervention trades some accuracy or robustness for fairness.ConceptAnchor-free detectionLocalizing objects or action segments without pre-set anchor windows: each location predicts its own boundaries, and a refinement step pools boundary features for each proposal. So far one card covers it (AFSD, temporal action localization). Its results are author-reported, and the official code differs from the paper in several settings.ConceptAngular margin losses (ArcFace and relatives)Train an embedding model through a classification head whose features and class weights are L2-normalized, so logits are scaled cosines, and penalize the true class by a margin on the angle or cosine. This forces tight, well-separated classes that transfer to unseen identities. Trade-off: the classifier matrix grows with the number of classes, and scale and margin must be tuned together.ConceptAnomaly detectionFlags inputs, pixels, sensor streams, graph nodes, robot actions or clinical decisions that depart from normal data, usually with few labelled failures. Covers defect maps, flow likelihoods, isolation forests, node contrast, OOD scores, streaming benchmarks, VLA failure detection and off-road navigation. Most numbers are author-reported, and no two papers share a setting.ConceptArithmetic intensityHow many floating-point operations an operation performs per byte it moves to or from memory. Low intensity means the chip waits on memory, not on arithmetic. Decoder Transformers at batch size 1 sit far below encoders on this measure, so they run slower despite similar FLOPs. Intensity depends on batch size, prompt length and precision, and a cut-off needs your own hardware numbers.ConceptAsymmetric quantizationLow-bit grids whose lowest level can sit away from zero, set by an offset next to the scale. On EfficientNet-B0 with 4-bit activations, a learned offset lifts accuracy over a symmetric grid, and most of the gain comes at 2 bits. For LLM weights, a per-row min-max grid collapses at 3 bits without GPTQ's rounding. Two cards so far: CNN activations and LLM weights.ConceptAttention interpretabilityScoring inputs by attention weights or by gradient-weighted or propagated relevance, then testing faithfulness. Attention often agrees with human rationales but was less comprehensive than gradient or LIME in one benchmark. Substitution tests show whether attention does work in a given model. Path patching finds circuit heads. Critiques say attention is not a reliable explanation.ConceptAttention mechanismAttention computes a weighted average of values, with weights from scoring each key against a query. Variants differ in the scoring function, the normalization (softmax, sparse, learned or sigmoid gate), and where queries, keys and values come from. Soft softmax attention costs quadratic time in the number of positions.ConceptAttributed network embeddingLearns a vector per graph node from its links and its node features, so each vector reflects the attributes around it. Two families so far: random-walk skip-gram (linear cost, feature vectors reusable on a second graph) and a GCN trained with mutual-information terms for multiplex graphs. Evidence is two papers, author-reported; both released codes differ from their methods in places.ConceptAudio deepfake detectionDecide whether a speech clip came from a speech generator. One card so far, WaveFake: a dataset of generated speech with two baselines, a per-generator Gaussian mixture model (GMM) on LFCC or MFCC features and RawNet2 on raw audio. The main choice is the GMM, which the authors find more robust, or RawNet2, which does better on average. Single paper, author-reported results.ConceptAuto-Context Model (stacked refinement with previous predictions)Auto-context feeds a stage's full-volume prediction back in as an extra input channel for the next stage, so each stage sees context from the one before. Shown here for a two-stage 3D GAN that maps MRI to CT. Reported gains are only plotted, the released code omits it, and every extra stage needs predictions for all training subjects first.ConceptAutoencoderAn autoencoder compresses an input into a code and rebuilds the input from it, so the code keeps what matters. Variants change the bottleneck (sparse, variational, discrete), the input (denoising) or the use of the code: an anomaly score, a training-only shape prior, a survival-model input, a generator latent or interpretable features. Choose by what the code is for and how many labels you have.ConceptAutomated fact-checkingChecking a claim against retrieved evidence and reporting the result with sources. Two published routes: an LLM with live web search that writes sourced replies to public posts, and a retrieve, select-rationale, label pipeline over fixed scientific abstracts. Author-reported results only. Trade-off: coverage and speed against evidence quality, cost and domain scope.ConceptAutomated interpretabilityScoring how well a learned unit of a language model can be explained, without a human reading every example. An LLM explainer writes a description from top-activating text, an LLM simulator predicts activations from it, and the score is their correlation. One card so far: learned sparse features score higher than neurons and PCA or ICA directions, but the score depends on which LLMs are used.ConceptAutomated machine learningFinding a model and its settings for a new dataset with little human design. Auto-sklearn 2.0 searches pipelines under a time limit, using a warm start, budget halving and a learned policy selector. CellForge has LLM agents design the model itself, bounded by discussion rounds and repair attempts instead of a time limit. Pick by time and compute budget, and by how much repeatability matters.ConceptAutomatic differentiationTaking exact derivatives of a network's outputs with automatic differentiation (AD) and building them into the model or its loss. Two patterns: a learned free-energy potential whose derivatives give stress, so the first law holds by construction; and PINN residuals where AD slopes are mixed with neighbour-point differences so sparse samples still constrain the flow.ConceptAutoregressive image generationAutoregressive image generators turn an image into discrete tokens, or predict pixels directly, and generate with a transformer. Covers raster-order token models, pixel-level models, a text-conditioned single-stream model (DALL-E) and next-scale prediction (VAR), with reported results, a scaling study and the constraints behind each choice.ConceptBackdoor attacks (trojans) and their repairA backdoor hides a trigger in a model's training data, so triggered inputs get the attacker's label while clean inputs behave normally. A survey sorts attacks and defenses into one taxonomy with no experiments. One neuron-pruning method repairs a backdoored model after training, without fine-tuning. All numbers are author-reported, from two cards only.ConceptBackpropagation through timeGradient-based training of a recurrent network by unrolling it over time steps and backpropagating through them. The central variant for spiking networks replaces the zero or infinite derivative of the spike threshold with a surrogate on the backward pass. Its cost is stored history, which online approximations trade for accuracy and locality. One card so far; the page is thin.ConceptBaseline subtractionSubtracting a baseline, usually the expected reward at the same context, from a policy-gradient reward to cut variance without changing the expected gradient. In on-policy distillation the closed-form baseline is the negative reverse KL; a top-k version is cheaper. Author-reported gains over plain distillation.ConceptBatchNorm foldingAt inference, a BatchNorm layer that directly follows a convolution is an affine map per channel, so it can be merged into that conv's weights and bias and removed. This makes BN cheaper to run than LayerNorm and similar norms. The trade-off: it only works where BN sits next to a linear layer, it requires choosing BN when the architecture is designed, and folded weights can be harder to quantize.ConceptBayesian networkA Bayesian belief network, either as an anomaly model with parameters averaged over their posterior and fitted on similar past cases (one clinical paper, best AUC-ROC 0.81 on 100 cases), or as a dependency graph over a dataset's attributes (one fairness survey of 15 datasets). No code for the first; no variance reported.ConceptBayesian neural networkA neural network whose weights are random variables, so each prediction is an average over plausible weight settings and comes with a spread. Exact inference is intractable, so practical methods approximate it: variational inference, Hamiltonian Monte Carlo, MC dropout or the Gaussian-process limit. They trade accuracy, memory and compute. One survey card covers this slug.ConceptBayesian optimizationSequential model-based search for the best hyperparameters or architecture when each evaluation is expensive: fit a surrogate to the trials so far, pick the next trial with an acquisition function, train it, repeat. The surrogate is usually a Gaussian process, a random forest or a TPE model; the cost of the surrogate and the number of trials are the main trade-offs.ConceptBayesian quadratureBayesian quadrature (BQ) estimates an integral by putting a Gaussian-process prior on the integrand and returning a posterior mean and variance, so the samples can come from any distribution. Only one card so far: a Bayesian policy-gradient paper applies it to policy gradients and reports, author-reported, fewer samples than Monte-Carlo on small test tasks.ConceptBenchmark validityHow to tell whether a benchmark score means what it claims: run-to-run variance, pass^k reliability, dataset diversity and held-out refreshes, controlled negatives, comprehension gates, judge checks, control probes, scoring-rule details, scoped naming, proxy calibration. The variants suit different constraints; most cited results are author-reported.ConceptBezier curve parameterizationPredicts an open curve, such as a lane, as a few learned control points and evaluates a Bezier curve through them to get dense points. Compact and smooth by construction; the trade-off is a fixed control-point count. The wiki has one card so far.ConceptBibliometric analysisMeasuring what a research literature contains over time: which languages, methods, datasets, tools or team types appear in which venues, and how central or novel they become. Four variants: venue-level inclusion over resource classes, entity co-occurrence with a z-score, entity-pair novelty, and descriptive counts over database records. Each depends on how records are detected and merged.ConceptBilevel optimizationNested optimization where an outer variable is scored by an inner problem's solution. Penalty-based first-order methods avoid Hessians; with heavy-tailed gradient noise, clipped inner loops and a normalized outer step give a stated complexity bound.ConceptBilinear poolingFuse two feature maps by summing their channel-wise outer products over spatial positions, then signed square root and L2 normalisation. Compact bilinear pooling keeps the output at a fixed size (16384 in RGB-N). Only one paper card so far, which fuses an RGB and a noise stream per region; no comparison with other fusion methods, runtime or memory numbers are available here.ConceptBinary neural networkNetworks whose weights and activations are approximated by sums of several binary terms, so that convolutions become bitwise operations. ABC-Net reports 65.0% ImageNet Top-1 for ResNet-18 with five weight and five activation bases, against 69.3% full precision and 51.2% for XNOR-Net. Each layer costs M×N binary convolutions, and no speed is measured.ConceptBipartite matching for set-prediction lossesTrains a model that outputs an unordered set by pairing each ground-truth element with one prediction via a minimum-cost assignment (Hungarian algorithm), then applying the loss to those pairs; unmatched slots learn "no object". It replaces anchors and NMS in detection, at the price of slow convergence and a fixed cap on outputs. Variants add one-to-many supervision during training.ConceptBird's-eye-view (BEV) representationA top-down grid of the scene that 3D detectors read and driving world models roll forward. Detector designs differ in how camera features reach it: lifted with predicted depth, written by camera attention, cropped per proposal, or kept as interacting maps. Trade-offs: compute, calibration and sensors. Numbers are author-reported.ConceptBird's-eye-view perceptionFill one top-down grid around the vehicle from cameras, LiDAR or shared agent features, then read map elements, 3D objects, lane links or off-road costs from it. Variants: HDMapNet, BEVFormer, V2X-ViT, TopoLogic, TopoFG, TopoPoint, SALON, multi-task BEV (2606.02979). Author-reported on different datasets.ConceptBootstrap confidence intervals for benchmark resultsResamples the runs, data splits or tasks behind a benchmark score to get an interval, instead of quoting one number. Three papers use it: paired runs for an A-versus-B decision, stratified runs for RL aggregates, and tasks for agent pass rates. All are author-reported, and the first two disagree on how far to trust the decision thresholds.ConceptBounding-box regression lossThe loss that trains a detector's predicted box against its target box, built on IoU with optional penalty terms. Three wiki cards: Inner-IoU swaps the IoU term for the IoU of scaled auxiliary boxes; Edge YOLO uses plain CIoU; a loss survey gives the GIoU formula. Use a penalised IoU loss by default. The Inner-IoU ratio needs tuning, and its gains are single runs.ConceptBradley-Terry modelThe Bradley-Terry model turns pairwise win, loss or tie outcomes into one latent strength per item. Covers the plain fit, tie handling, a task-aware variant fit by EM for ranking robot policies from real-robot A/B trials, and a reward-model loss trained on preferred clips. Based on three papers so far.ConceptBranch-and-boundExact search for mixed-integer programs: split one fractional variable into two child problems and prune subtrees that cannot beat the incumbent. The branching rule, which variable to split on, decides the tree size. Shown here with strong branching, SCIP's reliability pseudocost default and a learned graph-network rule, on four NP-hard families (author-reported).ConceptBundle adjustmentBundle block adjustment (BBA) jointly refines camera poses, calibration and tie points against ground control points. Its settings (GCP count and accuracy, camera-position source, self-calibration, tie-point accuracy and filtering, post-transform) change check-point error by an order of magnitude in a 768-variant UAV study with Agisoft Metashape. Based on one paper so far.ConceptCalibrationWhether a model's stated uncertainty or confidence matches how often it is right. Measured with ECE, RMS calibration error or NLL, or set as a conformal threshold for a success rate. Sources are author-reported and use different binning and settings, so their numbers are not directly comparable. SNGP, MC dropout, ensembles and GPR differ in over- and underconfidence.ConceptCamera pose encodingHow a transformer is told where each camera is and how its lens bends light. A per-token ray frame handles any lens shape and, with a gravity-aligned map, gives absolute pitch and roll for video. Plücker ray tokens serve 3D reconstruction. Each card reports numbers only in its own setting, so the two are not directly comparable.ConceptCanny edge detectionCanny edge detection: a multi-stage gradient edge detector, here followed by a Hessian ridge filter. One source (2609.07408): as preprocessing for small tabular classifiers on spiral drawings, the Canny plus Hessian variant lowered AUC for 8 of 9 classifiers, and augmentation helped most. Author-reported, single-source page.ConceptCarbon accounting for AI computeHow to estimate the operational CO2 of a cloud GPU job: measured GPU energy times the grid's marginal carbon intensity at the time and place it runs. Covers the GPU-only scope, the intensity data needed, and why the estimate is a lower bound.ConceptCarbon-aware schedulingChoosing where and when to run a GPU job so that it emits less CO2, using the grid's marginal carbon intensity. Three levers: region choice, Flexible Start and Pause and Resume. A 2022 Azure study finds region matters most. Its gains come from simulations on 2020 historical data with perfect foresight, so they are upper bounds.ConceptCatastrophic forgettingA trained network loses earlier abilities when it is trained on new data. This page covers the settings studied, two ways of measuring the loss that do not give comparable numbers, replay with a small memory, distillation toward a previous model, and keeping a frozen model's outputs close during prompt tuning. Other methods are on continual-learning.ConceptCausal discoveryLearning which variables have a direct causal effect on which others, from observational data and optional metadata. One source so far: a multi-agent LLM debate that picks a statistical algorithm, runs it, then refines the graph with metadata. Author-reported wins on five benchmarks; 16 to 25 times the tokens of standard baselines. Trade-off: accuracy against token cost.ConceptCausal inferenceA 2021 review finds counterfactual methods ungrounded. A 2021 NLP survey lists text-effect assumptions. A 2022 survey gives the SCM core. A 2020 review says importance scores need assumptions. A 2022 GPT-3 study is mixed on interventions. A 2023 paper finds LLM subjects confounded. A 2024 trial separates access from use. A 2026 method replays agent steps. Trade-off: what-if answers vs grounding.ConceptCentralised training with decentralised execution (CTDE)Cooperative multi-agent learning where training may use global state, but each agent acts only on its own history at run time. One card compares a centralised critic, joint-value methods and independent PPO on StarCraft maps. Independent PPO matched or beat the others on several hard maps in author-reported runs. Pick by the state available in training and the coordination risk you accept.ConceptChain of thoughtMake a model write intermediate text before it acts or answers. As a training target, grounded chain of thought adds a subtask, a movement plan and a gripper target before each robot action. On an edge GPU, long chains dominate latency: token caps, parallel samples and quantization trade accuracy for speed, and pruning low-entropy steps cuts thinking tokens by 27-45%.ConceptChain-of-thought compressionReplace the lowest-entropy steps of a reasoning trace with [SKIP] markers, on existing traces or as a behaviour learned with SFT and GRPO. Author-reported: static pruning cuts thinking tokens 27-45% on MATH500 and AIME, but AIME 2024 accuracy falls 6.7-6.9 points on two DeepSeek models. The learned variant needs a large SFT and RL budget.ConceptChain-of-thought promptingMake a language model write out intermediate reasoning steps before its final answer, using worked few-shot exemplars or a zero-shot trigger phrase. PaLM 540B goes from 17% to 54% on GSM8K with 8-shot CoT (58% with a calculator). Sampling many chains and voting adds more, at several times the decoding cost. CoT can also lower accuracy on some commonsense tasks.ConceptChallenge-set evaluationProbes one known weakness with a small set whose items isolate it, reading one property of each output. Variants: translation gender bias, hand-built QA with an "unknown" option, minimal edits, image-caption minimal pairs, sampled completions, curated failure archive, test suites, hard NLI subset. Each needs its own reading rule; API results may not reproduce.ConceptChamfer distanceChamfer distance measures how far two unordered point sets are from each other by averaging nearest-neighbour distances. Covers the squared-L2 form used for training and evaluation (GRNet), a density-aware variant for measuring deformation (SoGraB), a plain form used to score generated objects (OLiDM), an L1 appendix variant, and a grid loss that is not Chamfer.ConceptChannel attentionChannel attention rescales a feature map's channels with a learned gate, usually squeeze-and-excitation. Seventeen papers cover it, mostly placing the gate in residual, fused, unfolded, nested, decoder, denoising, batch-norm, detector, spiking or two-stream blocks. Removing DNANet's gate costs 0.82 IoU on NUDT-SIRST.ConceptChart data extractionRecovering a data table of values and labels from a chart image, so that code or a model can read the numbers. The core idea: detect marks and text, read the text, map mark positions to values through the axis scale, and pair values with their labels. The main trade-off: extraction errors lower downstream accuracy, for example from 61.84% with gold tables to 45.52% with extracted tables.ConceptClass-balanced lossLoss weights set per class from its training count, so rare classes count for more: an effective-number weight, or an inverse-frequency power. Covers classifier-only retraining, training from scratch and per-pixel uses, and where each helped or hurt. Three wiki cards use it; numbers are author-reported, and some card tables and code disagree.ConceptClassifier guidanceClassifier guidance steers a pretrained diffusion model toward a target by adding a differentiable score's gradient during sampling, without retraining. Covers four forms (sampler shift, cross-modal mismatch, point estimate under a constraint, discrete logit adjustment), their costs, and scale pitfalls.ConceptClassifier-free guidanceClassifier-free guidance (CFG) steers a generative sampler toward a condition by combining a conditional and an unconditional prediction at every step. It needs training with the condition sometimes dropped. Covers weight conventions, negative prompting, rescaling, per-condition weights, blending two models, logit guidance for token models, training-time dropout, and physics-residual conditions.ConceptClustered federated learningClustered federated learning groups clients or their data before averaging, so each group shares one kind of data. A COVID-19 study clusters clinics by imaging modality, with clusters set by hand; its results are close to per-modality models. k-FED clusters data split across devices in one round, with local k-means and a server step. No source here reports measured bandwidth or privacy.ConceptClustering evaluationScoring a found clustering against reference labels with external agreement indices: pair-counting indices (Jaccard, ARI, Fowlkes-Mallows) and information-based ones (NMI, AMI). Pair counts compare which objects share a group; the indices give different numbers on the same results, so report which one you used.ConceptCode completionSuggesting code while typing, and chat help with coding tasks. Covered by Code Llama and StarCoder (infilling, benchmark pass rates), CodeGeeX (13B model with editor plug-ins), phi-1 (1.3B model on curated data) and a student study of Copilot and ChatGPT. No head-to-head comparison exists, so no tool is ranked. Numbers are author-reported.ConceptCollaborative filteringPredicts a user's rating of an item from past ratings by learning a vector per user and per item. Rating-only methods struggle when users and items have few ratings. Adding review text helps: DeepCoNN reports 7.6% to 8.7% lower MSE than the best baseline on Yelp, Amazon and Beer.ConceptCollaborative perceptionConnected vehicles share what they sense, so each sees objects it would miss alone. Variants differ in what crosses the link: detections (late), raw point clouds (early), compressed features (intermediate) or object queries (UniV2X). Author-reported: intermediate fusion led on OPV2V and on V2V4Real at AP@0.5; under pose and delay noise on V2XSet, late and early fusion fell below no fusion.ConceptCommonsense knowledge injectionCommonsense knowledge injection gives a task model knowledge from outside its weights, generated or retrieved per input. COSMIC adds COMET relation vectors to emotion states and CEM to reply encoders; QA-GNN and GreaseLM fuse a retrieved ConceptNet subgraph with a pretrained LM. Trade-off: extra per-input work, for author-reported gains from under 1 to about 5 points.ConceptCommunity detectionSplitting a graph's nodes into groups whose members link more among themselves. Covers stochastic block models (degree correction, mixed membership, nested), modularity, fitting and choosing the number of groups, and Leiden communities used by GraphRAG. Review figures only; GraphRAG results are author-reported.ConceptCompartmental epidemic modelCompartmental models sort a population or network into states such as susceptible, infected and recovered, and move members between them at timed or rate-based transitions. Three cards: an agent-based SEIR model, an SIR form with an awareness factor, and an SIRD model with per-node risk for traffic-network failures. None is validated against observed outcomes.ConceptComputation offloadingDeciding, per time slot, how much of a device's work to send to nearby base stations instead of running it locally, so that long-run energy stays low and queues stay stable. One source uses Lyapunov optimisation with an actor-critic agent, evaluated only in simulation. A TinyML review argues for keeping inference on the device instead.ConceptConditional batch normalizationConditional batch norm normalizes features as usual, then scales and shifts them with values that depend on a conditioning input such as a sentence. The only card here is SD-GAN's semantic-conditioned batch norm, with a sentence cue or a word cue. Author-reported gains come only from ablations on a text-to-image GAN, and its cue-level numbers disagree across tables.ConceptConditional random fieldA CRF scores a whole label sequence from emission and transition terms and decodes with Viterbi. On BERT it beats a softmax head by only 0.35 F1 (LAPTOP) and 0.84 (REST). A neural CRF beats a log-linear CRF on high-resource targets, and on low-resource ones only with a related source language. Dense depth models reuse the name for windowed refinement.ConceptConformal predictionTurns model scores or errors into sets, alarm thresholds or safety levels at a chosen error level, from labelled calibration data and one quantile. Shown for LLM planners (a singleton set means act), VLA failure alarms, classifier credal summaries and a quadruped value-function filter. Assumes exchangeable calibration data. Four author-reported papers.ConceptConsistency regularizationA loss asking a model to give the same prediction for an input and a perturbed copy, or for an input and a frozen reference model. Variants differ in what they perturb (noise, adversarial direction, augmentation, dropout, weights) and where the target comes from (same pass, past epochs, weight average, pseudo-labels, frozen model). Trade-off: extra passes or stored state versus target freshness.ConceptConstrained conditional modelsStructured prediction by training several simple local classifiers independently and combining their scores at inference time, under hard constraints on the output (e.g. a positive answer, an integer answer) enforced by a beam search. Needs little jointly annotated data; the trade-offs are a heuristic search and a constraint check that can silently fall back to an infeasible-looking output.ConceptConstrained decodingSteers a model's output during decoding. Grammar masks keep each token inside a formal grammar but need next-token probabilities. Input-token masks restrict output to tokens that appear in the input. Energy-based COLD scores whole drafts against weighted constraints, with no guarantee. Author-reported: 100% valid constituency trees with per-input grammars; 94.50% keyword coverage on CommonGen.ConceptConstraint embeddingHandling closed kinematic loops in a mechanism by solving each loop's constraints numerically inside its own limb, so that the limb acts as an ordinary serial chain in independent coordinates. The trade-off: the local solver needs well-conditioned loop Jacobians and regular, non-redundant constraints, and the only evidence so far is one author-reported two-DOF example.ConceptConstraint satisfactionModel finite-domain problems as variables, domains and rules, then let a solver find assignments that meet every rule, or optimise one quantity. Covers the Python-embedded route that compiles to the XCSP3 format and the choice of solver backend. One source so far: a library guide with no benchmarks.ConceptContext distillationContext distillation trains a model, without the document in its prompt, to imitate its own answers made with the document present, so the document ends up in the weights. Per-document training is slow. A hypernetwork can write the adapter in one forward pass, under a second in the paper's runs. Trade-off: a one-time hypernetwork training run and lower reported quality than per-document training.ConceptContinual learningLearning new tasks or classes without forgetting old ones. Variants: gradient projection (A-GEM), tiny replay (ER) or frozen columns (progressive nets) for tasks; exemplars (PODNet, ORE), class means (SDC) or expansion (DER) for classes; pseudo-labels (PLOP); not-true distillation (FedNTD); weight blending (PAINT); expert freezing (SAME). Trade-off: stored old data against per-step cost.ConceptContinuous convolutionA convolution defined over continuous coordinates, so it runs on unevenly spaced points such as lidar returns and can output at any location. The kernel is a small MLP of the relative offset between points. One card so far, reporting gains on indoor segmentation and lidar motion; the reported numbers are author-reported and the training details are not stated.ConceptContinuous text diffusionRuns diffusion on continuous word-embedding vectors instead of tokens, so each step sees the whole sequence. Covers Diffusion-LM (x_0 prediction, gradient control of latents), LangFlow (cross-entropy, learned noise schedule) and S-FLM (unit-sphere flow). Costs: many sampling steps, and weaker likelihood or accuracy than autoregressive models.ConceptContinuous-time state estimationEstimates a state that is defined at every time instant, not only at sensor timestamps, so it can be queried between samples. A sliding window of recent time steps keeps the cost bounded for live use. One paper so far: a 0.1 s window on a continuum robot matched or beat a 10 s batch smoother on all five trajectories, at 8.7 to 10.2 ms per step (author-reported).ConceptContrastive learningTrain an encoder so queries land near their positive and far from negatives, usually via an InfoNCE softmax over one positive and K negatives. Variants differ in where negatives come from (batch, memory bank, queue), whether they are needed (BYOL-style), what counts as positive (augmentations, a patient, a class), and softmax vs per-pair sigmoid. Batch size usually decides.ConceptControl barrier function (CBF)A barrier function defines a safe set as a superlevel set, and a controller keeps the state inside it. Backup CBF builds the safe set from a backup policy and filters the nominal input with a quadratic program. Learned CBFs in a differentiable QP layer, as in LiveNet, predict the barrier penalties with a network. Two sources, both with simulation results only.ConceptControllable text generationControllable text generation makes a model's output meet a condition, such as a parse tree, a keyword or a speaker's gender. Seven method cards cover seven routes, from diffusion latent steering and frozen-model energy decoding to a moral-direction filter and control tokens, and a survey card maps the field. Results are author-reported.ConceptControlled natural languageA controlled natural language (CNL) is an everyday language, such as English, with deliberate restrictions for readability, translation or formal use. A survey of 100 English-based CNLs scores each on the PENS scale: precision, expressiveness, naturalness and simplicity. Use it to find prior designs and to place a trade-off. It has no method or experiments.ConceptConversational question answeringAnswering the next question in a chat about a passage, using earlier turns. One paper (CoQA) gives a dataset and three baselines: a seq2seq copy model, a span reader, and a pipeline of both. Best reported test F1 is 65.4 against 88.8 for humans (author-reported). Trade-off: span readers cannot give free-form answers, and the best recipe is not in the released code.ConceptConversational recommendationDialog systems that recommend items during a chat, such as movies. One paper with a card labels each recommender turn with a social strategy (opinion, similarity, encouragement, credibility) and conditions a GPT-2 model on it. Author-reported gains on perplexity, BLEU and human ratings, on a small dataset; the link to success is correlational. Trade-off: needs strategy-annotated dialogues.ConceptConvolutional neural networkA network that slides learned filters over local neighbourhoods, stacking convolution, nonlinearity and down-sampling. Variants differ in dimension, depth, skips, recurrence, branching, sparsity, regularisation and hybrid transformer blocks. The 71 papers span segmentation, registration, saliency, text, pruning, transfer, video, point clouds, grasping, mineral mapping and robot force signals.ConceptConvolutional recurrent network (CRNN)A convolutional front end feeds a recurrent layer, or convolutions sit inside the recurrent cell. The papers here put the recurrence on different axes: time for music tagging and satellite series, image width for text reading, unrolled steps for histopathology, and stacked memory across layers and time for video prediction. Every result is author-reported.ConceptCooperative driving automationConnected automated cars share status and intent to agree on joint moves such as platooning or merging. Test benches couple vehicle and traffic simulators. Variants differ in test setup, join rule and following law. The evidence so far is one simulation-only freeway study.ConceptCoordinate classificationPredicting each keypoint as one distribution over discrete bins per image axis, instead of a 2-D heat map. The paper's variant (SimCC-style head with one gated attention unit) runs fast on CPU, GPU and mobile at 256x192 input. Main trade-off: a discrete grid and a soft-label loss to tune, against heat-map upsampling cost. One card so far; author-reported.ConceptCosserat rod theoryA Cosserat rod models a slender body as a centreline with a rotating cross-section, and gives force and moment balances. Variants differ in the solver (boundary-value methods, or energetic Ritz reductions such as GE-FEM, modal Ritz and GVS) and in whether the cross-section may change size. Trade-offs: a consistent model for beam-like soft robots, but hard to implement and to reduce for control.ConceptCost aggregationRefining a cost volume, a table of scores over candidate matches, before a mask or a disparity is read out. Compared: 4D convolutional and transformer aggregators for few-shot segmentation, and a 3D hourglass network for stereo. Main trade-off: memory and run-time of the volume against global context. Author-reported.ConceptCost volumeMatching scores between views, one per candidate depth or disparity, or per pixel pair for flow. Variants: group-wise and depth-indexed stereo volumes, plane-sweep for posed views, all-pairs latent tokens for flow, and a parameter-free softmax matcher shared across tasks. Trade-off: explicit matching evidence against memory that grows with the candidates. Author-reported.ConceptCounterfactual explanationA counterfactual explanation names the smallest change to an input that flips a model's prediction, such as "the loan would have been approved with a higher income". Tabular methods differ in search mechanism, constraints and causal grounding; NLP work edits text minimally. Six reuse-licensed cards cover it so far; the food-quality review names it only as an open problem.ConceptCox proportional hazardsNeural versions of the Cox survival model: a neural risk term trained on a sampled-control loss, a Cox partial-likelihood term added to an autoencoder loss, or time as an input that drops proportional hazards. Author-reported gains on KKBox churn data and small survival sets; the autoencoder paper has one centre and internal validation only.ConceptCredit assignmentDeciding which steps, agents or inputs deserve credit or blame for an outcome. Covered: learned reward redistribution that splits an episodic team return into per-step, per-agent rewards; per-action advantages for LLM agents; intervention-based attribution of failed agent runs; and Shapley-style fair shares. Evidence is mostly author-reported, with synthetic-only and survey sources.ConceptCurriculum learningSchedules training so that simpler examples or easier targets come first. Most order-based variants in a 31-team small-data shared task did not beat a random order. Other variants schedule task difficulty, a pseudo-label threshold, a loss weight, an input source or a staged training pipeline instead. Test against a fixed-schedule baseline before adopting one.ConceptDDIM inversionRecovers the noise latent a deterministic DDIM sampler would turn into a given image, so the noise can be resampled under new conditions. Plug-and-Play inverts with an empty prompt and about 1000 steps; Composer uses the image's own conditions; StyleAligned inverts one style reference under a caption. Trade-offs: cost, and inversion can fail.ConceptData augmentationExtra training signal from labelled data: transform examples while keeping labels valid, mix patches with label weights, generate images, or label recombined pairs with a teacher. Choose by label type, scarcity and what you can change.ConceptData poisoning (attacks on training data and sanitization defenses)Training data can be poisoned on purpose, or perturbed by owners to resist learning. Clean-label polytope poisons transfer to unseen victims (author-reported). Outlier sanitization is the classic defense, and one survey finds none held up. LLM-written fakes in a retrieval corpus cut QA exact match by up to 87% (author-reported).ConceptData portabilityHow a person can move or reuse data that a service holds about them, and the three architectures that set who controls the data and how it moves. Based on one legal analysis of EU rules (GDPR, Data Act, DMA) with no experiments, so no measured results.ConceptData valuationScoring each training data point by how much it contributes to a model's performance, most often with the Shapley value. Covers the cooperative-game setup, exact and sampling-based computation, and the cost and axiom trade-offs. Based on one survey with no experiments, so the numbers are cited claims, not measurements.ConceptData-consistency layerA step inside learned image reconstruction that pulls the output back toward the measured samples. As a layer, it sets sampled k-space entries to the measurements (noise-free) or mixes them with a weight λ (noisy). As a GAN loss, a cyclic term plays the same role. Shown for cardiac and compressed-sensing MRI.ConceptDataflow mappingDeciding which operands stay on chip and get reused, and in what order work runs, in a DNN accelerator. Four dataflow styles (weight-, output-, row-stationary, no-local-reuse) trade control simplicity against buffer bandwidth. MAESTRO and Timeloop estimate a mapping before hardware exists. The only source is one FPGA survey that quotes numbers from other works.ConceptDataset curationHow training and evaluation data is sourced, filtered, deduplicated, labelled and documented. Covers web text, image-text pairs, speech, vision, 3D objects, parallel text, robot demonstrations, instruction and benchmark data, the filters each paper used and their reported effects, and the licence and documentation traps that come with reusing them.ConceptDebiasing word embeddingsRemoves gender bias from word vectors by shrinking their projection on the he-she direction. The paper tested two methods, Hard-Debiased and GN-GloVe, and found that gender stays recoverable from neighbours, clusters and a classifier. Judge a debiasing step with several checks, not the projection score alone.ConceptDecision-focused learningTraining a predictor on the regret of the decision a downstream solver returns, not on its prediction error. The solver has zero gradient almost everywhere, so each method substitutes a surrogate: differentiated optimality conditions, smoothing, random perturbation, or solver-free losses over cached solutions. Trade-offs: solver calls per step, constraint types, and how tight the LP relaxation is.ConceptDecoding strategiesHow a model picks each next step from its scores: greedy, temperature, top-k and top-p sampling, CoT-decoding, majority voting, beam search, contrastive search, contrastive decoding, MBR over robot action chunks, toxicity filtering and tree-based speculative verification. Twenty paper cards; numbers are author-reported.ConceptDeep Q-learningLearning action values with a parametric function instead of a table, usually with experience replay and a target copy. Six cards cover quantum, driving-test, driving-reward, robot and hardware-survey variants. Results are author-reported and do not compare variants on one task.ConceptDeep canonical correlation analysisDeep CCA trains two networks so that their outputs are as correlated as possible, with a loss computed over each batch. The paper's ICCN applies it to text-based audio and video interaction matrices for sentiment and emotion, and reports gains over text-only models. Its released code departs from the paper in several places, so the reported numbers are hard to reproduce.ConceptDeep learningMachine learning with neural networks of many stacked layers, which learn features from raw data instead of hand-designed ones. Covers three modelling paradigms, architectures by data type, deployment on microcontrollers, and when to prefer shallow models or transfer learning. Six sources; the library and microcontroller papers report their own experiments.ConceptDeep supervisionAttaches losses to intermediate predictions as well as to the final output. A saliency paper puts losses on side outputs at three encoder depths and reports lower TORONTO scores without them. A 3D instance-segmentation paper applies a matched loss at every decoder layer. Two cards so far.ConceptDeep unfoldingDeep unfolding turns a fixed number of solver iterations into learned network stages, trained end to end. Variants differ in how the measurement enters each stage, whether stages share weights, and whether the data step is a gradient or a closed-form replacement. A known operator is simplest, and cost grows with the stage count. Cards: image restoration and dynamic MRI.ConceptDeformable convolutionA convolution whose sampling points move by learned offsets, so each output reads content-dependent positions. Five cards use it differently: a deformable first layer in detector heads, deformable sampling in 2D and 1D attention, a backbone operator (DCNv3), and flow-guided video alignment. Gains are author-reported and not comparable across papers.ConceptDenoising autoencoderTrain an encoder and decoder to rebuild a clean input from a corrupted copy, so the code keeps the structure needed to rebuild it. Variants differ in the corruption, what the decoder may read, and what the code feeds: a survival loss, a sentence vector, a shape penalty, sparse-input pre-training, or time-series image repair for imputation. Choose by input type and label count.ConceptDense correspondenceMatching pixels between images, or points on 3D objects and bodies. Covers label-free diffusion features (DIFT), one Transformer matcher for flow, stereo and depth (UniMatch), coarse-to-fine matching on frozen features (RoMa), GAN-feature tracking (DragGAN), descriptor fields for pose search (NDF) and body-object correspondence (BEHAVE). Trade-off: label-free convenience against per-task tuning.ConceptDense retrievalRanks documents by one query vector against precomputed document vectors, a nearest-neighbour lookup. Variants differ in what they train on or generate first: labelled search data, LLM-written queries or tasks, unlabelled LLM adaptation, conversation pairs, multilingual data, a generator trained jointly, LM-scored examples or reasoning text. Pick by labels, generation budget and serving size.ConceptDensity map regressionDensity map regression predicts a value per pixel whose sum over a region is the object count, trained against Gaussian-smoothed point annotations. One card covers it so far: DecideNet blends a detector's head boxes with a regression map through a learned per-pixel attention. The trade-off: the detector needs box labels and runs in every training step.ConceptDependency parsingDependency parsing gives each word of a sentence one head and a labelled relation to it, forming a tree. Four cards cover it: an easy-first parser in arithmetic word problems, the Universal Dependencies scheme, a biomedical toolkit with an arc-eager parser, and an arc-scoring parser on a French encoder. Only the toolkit card compares parser architectures, against one baseline.ConceptDependency treebank (annotated sentences with word-to-word dependency links)A treebank marks each sentence with a dependency tree: every word has one head and a relation label from a shared scheme. Covered: the basic single-head tree, the optional enhanced graph, the v1 to v2 changes a converter must apply, and a biomedical treebank converted to UD v1.0. Trade-off: one cross-lingual label set, versus the effort of adopting or converting it.ConceptDepth completionFills the holes in a sparse depth map, such as LIDAR, into a dense one. Unguided methods use only the depth map; guided methods also use a colour image. The one card here is a fixed chain of classical morphology and blur steps with no training data, about 90 Hz on a CPU, and the best KITTI test RMSE at submission. Its constants were tuned on KITTI.ConceptDepth separationDeep ReLU networks can reach error eps with polylog(1/eps) neurons when their depth grows like log(1/eps), while shallow networks need poly(1/eps) neurons for the same smooth functions. The result is a size-and-depth statement about what networks can represent, proved by construction and lower bound. It says nothing about training, and the constructions use hand-built weights.ConceptDepthwise separable convolutionSplits a regular k×k convolution into a per-channel k×k depthwise conv and a 1×1 pointwise conv that mixes channels, cutting FLOPs and parameters sharply. The catch: depthwise convs do little arithmetic per byte moved, so they run at low achieved FLOPS and the latency saving is often far smaller than the FLOPs saving. Partial convolution is one higher-FLOPS alternative.ConceptDialogue evaluationScoring dialogue systems without a human in every test. Task bots: task success and cost, with satisfaction models fitted to ratings. Chatbots: word-overlap, learned and adversarial scores, none reliable alone, or crowd labels. A 2019 survey finds weak agreement with human judges. Trade-off: cheap automatic scores versus validity and labelling cost.ConceptDialogue generationGenerating a reply from the conversation so far. Plain maximum-likelihood Seq2Seq gives generic replies; decoding changes, extra inputs (commonsense, strategy labels, database results), reward-based training and LLM adaptation try to improve on that. Results are author-reported on each paper's own data. Trade-off: reward-based training, labelled turns and LLM systems cost more.ConceptDice lossOne minus an overlap score between predicted and true masks, computed on probabilities. Five cards cover soft Dice, Dice added to cross-entropy or binary cross-entropy, and a soft-Jaccard loss that is a sibling, not Dice. Overlap alone can behave badly on small objects, so the cards pair it with cross-entropy. No card ablates the Dice term.ConceptDifferentiable architecture searchNeural architecture search by gradient descent: relax each edge's operation choice into a weighted mixture or one sampled path, train architecture weights on validation data, keep the top operation per edge. Search costs run from under one GPU-day on CIFAR-10 to about 200 GPU hours on ImageNet (author-reported). Full mixtures use memory in proportion to candidate count.ConceptDifferentiable fuzzy logicGives logical formulas truth values in [0, 1] and replaces AND, OR, NOT, implication and quantifiers with smooth operators, so a knowledge base becomes a training objective. One paper so far: Logic Tensor Networks, with stable product operators and p-mean quantifiers. Choices: operator family, quantifier exponents, masks, refutation. Results are author-reported.ConceptDifferentiable optimizationPasses gradients through a solver so a model can be trained on the regret of its decision or on a target control. Solutions of LPs and ILPs are piecewise constant in the costs, so each method swaps the zero derivative for a smoothing, a perturbation or a surrogate loss. The trade-offs are which problem classes a method supports and how many solver calls each training step costs.ConceptDifferentiable renderingDifferentiable rendering makes an image-forming model differentiable, so a loss on the image can update pose, scene, material, face or mesh parameters by gradient descent. Variants fit faces, silhouettes or photos, score renders with a frozen CLIP or diffusion model, grow point clouds, refine SLAM maps, shade night driving scenes, or separate light from material.ConceptDifferentiable simulationA simulator whose gradients drive a fit to data or a training loss. PhysDreamer fits a per-particle Young's modulus to a generated video. Physics-aware training takes gradients from a fitted digital model of a real device. TrafficSim unrolls a learned traffic policy in closed loop. Choose by whether you can simulate the system or must measure it. Results are author-reported, from three cards.ConceptDifferential privacyDifferential privacy (DP) bounds how much an analysis's output can reveal about any one record, stated as epsilon. DP-SGD (clipped, noised gradients) is the main training variant with reported settings. An audit shows measured leakage depends on attacker access and sits far below the proven bound for black-box attackers.ConceptDifferentially private SGD (DP-SGD)DP-SGD trains with per-example gradient clipping and Gaussian noise, and its proof bounds what one training example can reveal. Two cards cover it. One finds that black-box attackers measure far less leakage than the proof allows. The other finds that DP at epsilon 8 still leaks some personal data from language models. Neither sets a privacy level.ConceptDiffusion inpaintingFixing part of a diffusion or flow sample to known values, or training a model to take that part as input, so the rest is filled in. Used to steer human motion, join motion segments, keep robot action chunks continuous, and fill or insert image objects. Variants differ in training, mask or delay handling. Six cards, author-reported results.ConceptDiffusion modelsDiffusion models learn to predict the noise added to a sample and run that prediction in reverse, starting from noise. Variants differ in the space they work in (pixels, latents, tokens, graphs, trajectories, speech, motion, robot actions), how conditions enter, and the number of sampling steps. Ninety-nine cards, with author-reported results.ConceptDigital twinA software copy of a physical system or process, used to predict, test options and in some designs to control. Five uses in the sources: training a navigation policy, a six-level ship scale, edge replicas of users, a fluid-solver surrogate and an arm model for contact force. Only the arm model runs on a real robot, and only to compute torques.ConceptDijkstra's algorithmShortest-path search from one start node over a graph with non-negative edge weights, giving the distance to every reachable node. Used as the global planner on a static warehouse layout, with a PPO controller on top; the one card that uses it reports better scores than DQN and A*, from a single table with no statistics.ConceptDilated convolutionA convolution whose kernel taps are spaced apart by a dilation rate, used to keep segmentation feature maps sharp. One card (FastFCN) compares a dilated backbone at output stride 8 with a stride-32 backbone plus a module of parallel dilated separable convolutions. Gains are author-reported on PASCAL-Context; one source, so nothing is compared across papers.ConceptDimensionality reductionMaps high-dimensional points to fewer dimensions, usually 2-D for plots, keeping their structure. Methods trade local neighbourhoods against global layout. PaCMAP keeps both and is reported faster than TriMap on large data. PCA, ICA and MNF are linear options. Feature selection keeps original inputs instead.ConceptDirect collocationTurns a continuous-time vehicle model into algebraic constraints by approximating the state inside each sampling period with polynomials that meet the model at a few collocation points. One reuse-licensed card covers it so far: a tractor-trailer NMPC that uses three points per 0.1 s period. The node placement and the constraint form are not stated, so no variants can be compared yet.ConceptDirect multi-step forecastingPredicts all T future steps of a series in one model call, instead of feeding each prediction back as the next input. A single linear layer with this output beat published Transformer forecasters on nine benchmarks in one author-reported study. Variants add a trend split or a last-value anchor. The evidence is one paper, and its baselines used a different look-back.ConceptDirect set predictionFraming a task whose answer is an unordered, variable-size collection (e.g. objects in an image) as predicting the whole set at once: a fixed number of output slots, padded with "no object", trained with a permutation-invariant matching loss. DETR drops anchors and NMS; it costs a hard output cap and long training. Variants shorten training, add training-only supervision or change matching cost.ConceptDiscrete choice modelingPredicting which option a chooser picks from an offered set. Parametric random-utility models fix the utility form; RUMnet learns it with a neural network and can approximate any random-utility model. Author-reported on two datasets: best accuracy, but a random forest had the lowest Swissmetro loss.ConceptDiscrete diffusionDiffusion over discrete variables such as tokens, graph classes, layout attributes or draft tokens. Corruption masks or replaces classes; a network predicts clean classes and sampling runs back to clean data. Covers masked text, graph, layout, watermarking, block drafting and RL variants, plus a continuous comparison flow. Seven cards.ConceptDiscrete speech unitsTurns continuous speech into a short sequence of integer codes that a language model can read and write like text. Two routes are covered: k-means clusters over self-supervised speech features (the SpeechGPT tokenizer), and learned vector quantisers such as wav2vec 2.0's codebook. A survey of self-supervised speech describes both; no paper here compares them on the same task.ConceptDiscriminability index (d')A d' value measures how well two classes can be told apart, from their distributions alone. It covers the Bayes index d'_b, which is the optimal error rate for two normals with equal priors, and two cheaper indices, Simpson and Fitter's d'_a and Egan and Clarke's d'_e. The cheaper indices can understate separation when variances differ. Bayes error needs integration.ConceptDisentangled representationA disentangled representation gives each independent factor of variation its own part of the latent code. Unsupervised VAEs learn one with a penalty, but scoring needs known factors. One VAE study finds FactorVAE and beta-VAE scores track reasoning accuracy with few samples. MISA and an animation model split shared or source-shape codes from private or pose codes.ConceptDistributional alignmentMaking a model's answer distribution over a survey question match a human group's answer shares. Routes: prompt steering; fine-tuning on group response shares (lower Wasserstein distance, author-reported; needs survey data and GPUs); demographic backstories. Others score single-respondent agreement, test wording biases, or score a national survey benchmark.ConceptDisturbance observerAn observer that estimates the lumped disturbance on a robot joint, from model error, friction and external force, so the controller can cancel it. One card so far: a nonlinear disturbance observer added to impedance control for a two-robot rehabilitation teleoperation rig, with qualitative results only, author-reported.ConceptDomain adaptationReuse a model trained on a labeled source domain for a different target domain, usually without target labels. Variants adapt the data (translate or pseudo-label target samples), the loss, the features, the retriever or generator, the fine-tuning recipe, the merge or each test input. Each adds cost or fragility, and most gains are author-reported on narrow benchmarks.ConceptDomain generalizationTrain on one or more source domains and predict on an unseen target domain that is never used for training or model selection. Methods change the training data, the learned features, or the training strategy. Trade-off: most families add memory or compute, and the survey's benchmark shows small gains over plain ERM.ConceptDomain randomizationTrain under many random variants of a setting so the learned behaviour does not depend on one fixed version. Thirteen cards cover symmetry relabelling (Hanabi), physics, dynamics and terrain randomisation for legged, humanoid, hand and field robots, and Kubric scene generation. Numerical ranges are listed in five cards. All author-reported.ConceptDouble Q-learningChoosing the next action with one network and scoring that action with a second, target network, inside a bootstrapped Q-learning target. One card uses it: a Double DQN with prioritized replay that generates critical driving scenes (author-reported results). The card gives no comparison with plain DQN.ConceptDropoutRandomly zeroing units, weights or whole sub-networks during training to regularise a network. Covers unit dropout, DropConnect, feature-net dropout and variational dropout for recurrent layers. Trade-offs: the rate is sensitive and the papers define it differently, the gain over no regulariser can be small, and the scaling used at inference must match the training form.ConceptDual-number quaternionA dual-number quaternion r + ε s packs a rigid motion into two quaternions, with r unit-length and r·s = 0. Turning pose constraints into quadratic forms in (r, s) lets a penalised least-squares error be minimised with a trust-region method. Author-reported: for camera pose from points, it is far more accurate than the linear Faugeras-Toscani method under noise.ConceptDynamic graph embeddingDynamic graph embedding gives each node a vector that follows a graph changing over time, as snapshots or timed events. Covered: temporal random walks, warm-started autoencoders, temporal point processes and dynamic GNNs, from one survey taxonomy plus a general graph-learning survey. The trade-off: neither survey runs experiments, so no method is compared on numbers here.ConceptDynamic mode decompositionFits a low-rank linear model to snapshot data, giving spatial modes with a growth rate and frequency each, plus amplitudes that evolve in time. Variants differ in how the operator is computed and how the rank is chosen. The one card here uses a randomized SVD and an error-minimising rank search; its errors are in-sample.ConceptDynamic network expansionLearning new classes by adding capacity per step and freezing earlier blocks, so old knowledge is not overwritten. DER reports 72.41% average accuracy on CIFAR-100 with 50 steps unpruned and 72.05% pruned; a survey covers the wider family qualitatively. The trade-off is model size, which grows each step; the pruned variant is not released.ConceptDynamic neural radiance fieldA dynamic neural radiance field renders a moving scene from new views. One fits a monocular video with masks and predicts scene flow. Two others learn a field per tracked object in driving scenes, so objects can be moved or removed. All three need extra inputs. Training takes about a day for the first two; the third states no time.ConceptDynamic sparse trainingTraining a sparse network from a random start while its connections change: a fixed number of weights stays active, and drop-and-grow updates move the mask during training. The main choices are the growth rule, the per-layer sparsity distribution and the drop schedule. Only one card so far (RigL), so comparisons are against baselines the authors report.ConceptDynamic time warpingDistance between two time series that stretches or squeezes the time axis to align them, so a movement done faster or slower still matches. Used as the weak-learner distance in multi-class AdaBoost on facial-landmark sequences, and as a 1-NN baseline that ranks last of nine on a 112-problem benchmark. Trade-off: each comparison is a full alignment; window and normalisation are unstated.ConceptEfficient vision backbonesImage backbones designed for low latency on real hardware. Latency equals FLOPs divided by achieved FLOPS, so cutting FLOPs with memory-bound operators such as depthwise convolution often barely helps. Partial convolution (FasterNet) keeps FLOPS high by convolving only a quarter of the channels. Trade-off: speed rankings depend on the device, and FLOPs and parameter counts do not predict them.ConceptEmergent abilitiesHow to tell whether a capability or prompting technique appears only above some model size, and why small-model tests and scaling curves can miss it. Covers the scale and training-time axes, metric and baseline choices, and nine cards: two surveys, prompting, label, sampling, distillation, training-suite and analogy studies, and a framing report.ConceptEmotion recognition in conversationLabel each utterance's emotion in a dialogue. Three text designs: DialogueRNN (per-speaker GRU states, attention over history), DialogueGCN (windowed relational graph over a BiGRU) and COSMIC (speaker states plus COMET features), all taking speaker labels. HFusion adds audio and video. CEM's emotion head sits in a reply generator.ConceptEmpty space skippingEmpty space skipping evaluates no network samples where a precomputed map marks the space as empty. KiloNeRF builds an occupancy grid from a trained NeRF teacher and pairs it with early ray termination, reporting a 71x speed-up on Lego. DONeRF avoids the grid: a depth oracle places 2 to 16 samples per ray near likely surfaces, and needs training-time depth. Both results are author-reported.ConceptEnsemble clusteringEnsemble clustering fuses several base clusterings, each built from a randomised part of the data, into one consensus partition. FastMICE does this for multi-view data with anchor-based bipartite graphs. Author-reported: top average NMI rank on 22 datasets and near-linear cost. Trade-offs: the cluster count k must be given, and runs are random and not reproducible in the official code.ConceptEnsemble learningCombine several models into one predictor. Bagging and boosting over trees, stacking with a meta-model, and fusing the outputs of deep or LLM members are the recurring designs. Author-reported gains are often small or from one dataset. In an 11-dataset tabular benchmark, boosted trees were usually the strongest single model, and a mix with deep networks scored best.ConceptEnsemblesCombine several models, views of one input, or samples from one generative model. Gaussian-head ensembles also split uncertainty into aleatoric and epistemic parts. Fisher-weighted merging gives one model at one model's cost. Learned view weights beat the plain mean in most, not all, settings. Aligned diffusion samples help a little. An option-shuffled LLM vote is covered. All author-reported.ConceptEntropy minimizationAdds a loss that pushes predictions on unlabelled data toward low entropy. Author-reported: maximum squares 44.3 vs plain entropy 42.2 mIoU (GTA5 to Cityscapes); diffusion views add 0.7 to 2.9 points over TPT. Main trade-off: confident samples dominate the gradient. AdaContrast instead adds a batch-level diversity term.ConceptEquivariance constraintA training loss or network property that commutes with a known transform: transform the input and the output should follow. Keypoint and region detectors train it as a label-free loss under sampled warps. Molecule and graph models build it into the network: SE(3) for 3D atoms, permutations for graphs.ConceptError-state Kalman filterAn error-state Kalman filter propagates a nominal state with IMU readings and estimates a small error state on a manifold, so rotations stay singularity-free. Two LiDAR-inertial-visual systems use it: R2LIVE fuses LiDAR and camera residuals, R3LIVE adds photometric colour residuals. Accuracy figures are author-reported.ConceptEuclidean-equivariant networkA network whose outputs move with its input under rotation and translation, and in one design uniform scale, while routing and invariant features use only distances, speeds and norms. Keeps geometric features that transform with the scene and pattern features that do not. Three cards: EqMotion (motion prediction), NeuSE (object SLAM) and Neural Descriptor Fields (manipulation), author-reported.ConceptEvent-camera video reconstructionRebuilds colour video from one colour frame and the event stream after it, decoded at any time in the window. Learns per-pixel motion and colour from the events separately, then fuses them with flow. Author-reported PSNR: 25.40 dB on BS-ERGB gap 1, 28.68 dB on the authors' E2D2 set. Needs events aligned to the starting frame. One paper so far.ConceptEvidential deep learningA network outputs non-negative evidence per class, which becomes a Dirichlet with a built-in uncertainty value, so one forward pass yields a prediction and an uncertainty. Per-input opinions are fused with a reduced Dempster-Shafer rule so unsure inputs count less. Main trade-off: one paper so far, all results author-reported, and the released code differs from the paper in places.ConceptEvolutionary dynamics (sexual recombination as a learning rule)How sexual recombination and selection change allele frequencies over generations, and when that process matches a multiplicative-weights learning rule. Covers the two update orders (selection before or after recombination), the marginal-only equivalence, and what regret guarantees follow. Based on one paper so far.ConceptExpectation maximizationFitting a model with hidden assignments by alternating responsibilities (E-step) and parameter updates (M-step). Covers Gaussian-mixture EM over time-extended features for robot models, a gradient-step M-step for non-linear decoders, unrolled EM trained end to end, a recurrent variant, and latent-task-bucket EM for ranking robot policies. Three papers so far.ConceptExperience replayStoring past examples and mixing them into later training. RLVR: a correct-answer buffer with retirement (author-reported +3.0 over on-policy, one 7B model). Continual learning: a tiny per-class memory beats four baselines (author-reported). DQN and LLM pretraining: replay not isolated.ConceptExplainable AIMaking a model's behaviour understandable to the people who use or are affected by it. Covers interpretable models versus post-hoc explanations, global effect plots, feature attributions that can disagree or miss planted bugs, graph and LLM explainers, and choosing a technique by question. Eighteen cards; author-reported evidence, mixed on whether explanations help people.ConceptExplanation evaluation (judging whether an XAI explanation is good)How to check whether an explanation of a model is any good. A 2022 review splits "good" into twelve quality properties and catalogues 29 automated checks. Other papers use qualitative criteria, ablation curves, explainer disagreement, attention baselines or chain-of-thought prompt-change tests. No single metric is endorsed; the right check depends on what you can do to the model and data.ConceptExplicit feature grids (dense voxels and factorised planes)Stores a field's values in arrays read by interpolation: a dense voxel grid, or 2D planes covering each pair of input axes, fitted per scene or predicted per input. Plane memory grows with the plane count, not the grid volume, but planes couple through their products, so scales and initialisation interact. How the reads are combined and the decoder set the rest of the trade-off.ConceptExtractive question answeringAnswering a question by choosing a part of the passage instead of generating text. Span readers reach 52.6 test F1 on CoQA, or 65.4 with yes/no tokens. An attention-over-attention cloze reader reports 74.4 CNN test accuracy, single model. LinkBERT-large reports 81.0 average MRQA F1 against 78.5 for BERT-large. Author-reported, on different datasets.ConceptFace recognitionDeep face recognition maps each face crop to a fixed-length embedding and compares embeddings by cosine similarity, so unseen people can be verified or identified. Training is a classifier over many identities that is discarded at test time. Results depend on both the training data (breadth, pose and age variety, label noise) and the loss (plain softmax vs angular margin).ConceptFactorization machineA factorization machine scores a feature vector with a bias, per-feature weights and a factor dot product for every feature pair. On iPinYou it trails logistic regression in one paper (68.18% vs 68.81% pooled AUC) and beats it in another (75.52% vs 73.43%). Under tanh layers it reaches 70.70%. The CTR and DeepCoNN papers do not state a factor size.ConceptFair clusteringChooses k centers so that every large enough group of agents gets nearby centers in proportion to its size. The axiom is proportionally representative fairness (PRF); SEAR always satisfies it in polynomial time, with a 3 or 1+√2 approximation to proportional fairness. Theory only: no experiments, and standard k-means-type objectives can violate PRF.ConceptFarthest-point samplingGreedy selection of a spread-out subset: start from one point, then repeatedly add the point farthest from the chosen set. One paper (author-reported) applies it to the actions in replayed episodes to pick subgoal states for a goal-conditioned world model. The paper leaves the distance metric and start rule open, and its released code uses FPS for only one of several subgoal strategies.ConceptFault diagnosis (for multi-component systems)Finding faulty parts of a system made of many interacting components, then separating the root cause from its knock-on effects. Two sources: a rule-based framework for autonomous driving (qualitative runs only) and an LLM-agent benchmark on live Kubernetes faults (author-reported 38-73% diagnosis success).ConceptFeature attributionMethods that score each input element, or each spatial location of a feature map, by its share of a model output. Covers path-integrated gradient maps, Shapley values by sampling or by propagation, relevance maps for Transformers, class activation maps, counterfactual explanations, and whether rankings hold across models and agree between methods. Author-reported results only.ConceptFeature caching across timestepsSpeeds up diffusion sampling by running the full network only every N steps and reusing cached high-level features in between. Training-free, for U-Net samplers with skip connections. Author-reported 2.15x on Stable Diffusion v1.5 and 4.12x on ImageNet LDM-4-G, both with quality loss.ConceptFeature scalingRescaling each input column before training. Two benchmarks compared 5 and 12 scalers with no scaling, on 82 binary and 16 UCI datasets. The choice changed results for most models that react, and a poor one can score below no scaling. Tree ensembles barely react; SVMs, perceptrons and MLPs react strongly. A time-series model rescales each patch with past values only. All are author-reported.ConceptFeature selectionKeeping a subset of candidate input features: filter, wrapper (including swarm search), embedded (L1) and safe screening, which drops Lasso features that are provably zero before solving. Exhaustive search grows as 2^n. Reported speed-ups and accuracies are author-reported on different data and are not comparable.ConceptFeature upsamplingRaising a coarse feature map back to a finer resolution with a learned module instead of plain interpolation, so a dense-prediction head gets detail without a dilated backbone. One card (FastFCN's Joint Pyramid Upsampling) compares it with bilinear and FPN upsampling on PASCAL-Context. All numbers are author-reported, from one paper and one dataset.ConceptFeature whiteningTraining-time penalties that decorrelate the channels of a network's feature maps. One card so far: instance selective whitening penalises only the covariance entries that move under colour and blur change, aiming to remove style and keep content. Trade-off: extra training work and author-reported gains on one domain-generalization benchmark.ConceptFederated aggregationHow a federated server turns client updates into one global model. FedAvg weights clients by data share. FedNTD changes the local loss. Fed+ adds a per-client penalty without its update rule. FairFed reweights clients by fairness gap. FedDF distils client logits on unlabeled data. FedGen trains a server generator for data-free distillation. FEDL also averages gradients.ConceptFederated learningClients train on their own data and a server combines what they send, so raw data stays on devices. What a client sends sets the trade-off: weights (FedMA), class embeddings (FedProto) or distilled predictions (FedDF, FedNTD). FedDyn corrects each client's objective, FedGen trains a data-free generator, Ditto keeps a personal model per device, FairFed reweights for fairness, k-FED clusters once.ConceptFederated reinforcement learningReinforcement learning agents share parameters, gradients or reward signals instead of raw experience. Horizontal FRL fits agents that run the same task in separate environments; vertical FRL fits agents that share one environment but each see part of the state. The one source is a survey with no experiments, so the choice rests on the setting.ConceptFeed-forward reconstructionFeed-forward reconstruction predicts 3D from images in one network pass, with no per-scene fitting. Splatter Image, MVSplat, iLRM and FLARE output Gaussians; pixelNeRF is a radiance field; One-2-3-45 makes a mesh; PRN regresses face positions; ReconDrive and Forge4D handle driving and human video; GeFF encodes RGB-D into latent points. Results are author-reported.ConceptFew-shot example selectionChoosing which cases go into a few-shot prompt: a random or fixed set, the cases with the most reasoning steps, nearest-neighbour retrieval, a learned retriever, or cluster-based picks from unlabelled questions. Retrieval often scores higher but needs a labelled pool and a search per query. Gains are author-measured, on few tasks and models.ConceptFew-shot learningPredicting new labels from a few labelled examples. Routes: a frozen embedding with a linear classifier (79.64% 5-shot on miniImageNet; author-reported), in-context demonstrations, adapter, contrastive or prompt-vector tuning, a training-free CLIP cache, noisy-channel scoring, in-context segmentation, or LLM-written training data. Main trade-off: what can you train, and what does each query cost?ConceptFill-in-the-middle (infilling)Training a left-to-right code model to write the code between a given prefix and suffix. Covered here by one card: StarCoder applies character-level FIM at rate 0.5 in PSM or SPMv2 layout, and reports better single-line infilling than SantaCoder and InCoder. SPMv2 is not defined in the paper, and no other FIM card exists in this wiki yet.ConceptFinite element methodDiscretises a continuous body into elements and solves its equilibrium or transient equations numerically. Variants: full 3D continuum elements, geometrically exact elements for rods (GE-FEM), and FE as reference data for learned surrogates. Trade-offs: the most physically complete option, but costly and too heavy for real-time control.ConceptFinite-difference methodEstimating a derivative from function values at neighbouring points instead of the exact slope at one point. In physics-informed networks, stencils are applied to network outputs. Coupling a stencil with the automatic-differentiation slope at the point is the variant reported as more accurate with sparse samples, at higher cost per iteration. Numbers are author-reported.ConceptFinite-state machineA finite-state machine (FSM) switches between a few modes, and each switch is a threshold test on one scalar feature, so no classifier runs at decision time. Used for walk, sit and stair transitions on lower-limb exoskeletons. The thresholds can be personalized from user statistics or by Bayesian optimization on labelled transitions. Only one paper so far, with author-reported gains.ConceptFlow matchingTrains a network to move noise to data along a learned path, then samples by integrating it, usually in a few steps. Robot variants differ in chunking, auxiliary heads and horizon heads. Navigation runs a flow head asynchronously, and offline RL keeps the flow as a frozen prior. Language variants diffuse token embeddings, on a line or on a sphere. Choose by task, retraining budget and step cost.ConceptFocal lossCross-entropy scaled by (1 - p_t)^gamma, often with an alpha weight, so easy examples count less. Sixteen wiki cards use it across detection, heatmaps, tracking, segmentation, action localisation, scene graphs and set prediction. Few isolate its gain, and many leave alpha or gamma unstated.ConceptFoundation modelA model trained once on broad data, usually with self-supervision at scale, then reused for many tasks by prompting, adapting or building on its embeddings. The main choice is how to use it, given labels, compute and model access. The report's two central risks are emergence and homogenization.ConceptFourier featuresSine and cosine of scaled inputs, fed to a network next to or in place of its raw input, so it can represent fine detail, periodic structure or a low-dimensional condition. Variants differ in frequencies: several octaves for a diffusion denoiser, a period matched to the domain for PINNs, a random scale for a style field, or a box embedding for objects. Gains come only from each paper's ablations.ConceptFourier neural operatorMixes information across a whole grid by multiplying Fourier coefficients with learned weights, so each layer is a global convolution at quasi-linear cost. Two cards use it: as a vision token mixer (AFNO) and as spectral kernels inside a multiwavelet PDE operator (MWT). Variants differ in channel mixing, frequency selection and sparsity.ConceptFrame differenceSubtracting neighbouring frames, or features of neighbouring segments, to get a cheap motion signal for a 2D video model. Only one card so far: TDN uses pixel differences in early stages for short-term motion and feature differences as a gate in later stages for long-term motion. Author-reported gains are from ablations on Something-Something V1.ConceptFrequency regularizationHide the high-frequency bands of a sinusoidal input encoding at the start of training and reveal them step by step. Shown for few-shot NeRF: FreeNeRF reports 19.92 object PSNR on DTU with 3 views and 1.02-1.04x the baseline training time, author-reported. Main trade-off: the schedule end T, which the paper sets differently for 9 views in two places.ConceptFrequency-domain learningMove a signal into a Fourier or DCT basis and learn on the coefficients. Time-series: FreTS, a supervised forecaster; CoST, a contrastive encoder with a Fourier seasonal branch; FourierGNN, a spectral graph layer; ETSformer, top-K Fourier bases for seasonality. Image: DCT coefficients as an extra input for face forgery detection. Trade-off: released code departs from the equations.ConceptFréchet Inception Distance (FID)A set-level score for generated images: the Fréchet distance between Gaussian fits to pretrained Inception features of real and generated images. Covers the base score and its variants for spatial detail, classes, memorization, small samples, preprocessing, video and feature choice. The score depends on sample count, feature network and preprocessing.ConceptFunctional safety (for AI systems)Checking that an AI system works as claimed, before judging its ethics or trade-offs. Six sources: a failure taxonomy from 283 incidents, a three-layer framework for generative-AI safety evaluation, a corner-case description and a fault-diagnosis framework for automated driving, a formal-verification position paper, and a risk taxonomy built from 74 AI risk frameworks. None sets a threshold.ConceptFused attention kernelA fused attention kernel computes softmax(QK^T)V one tile at a time in fast on-chip memory, so the full score matrix never goes to slow GPU memory. Variants differ in how work overlaps on the GPU and in numeric precision. The evidence so far is one Hopper (H100) paper, with author-reported speed-ups and FP8 accuracy measured on synthetic data.ConceptFuzz testingFuzz testing searches for inputs that make a system fail by mutating candidates and keeping those that expose new failures. This page covers simulation-based fuzzing of driving software, where a candidate is a traffic scenario. One card so far: a genetic fuzzer with a learned violation score, author-reported to beat three baselines on one system and one map.ConceptGAN inversionFinds the GAN latent code that reproduces a real image. Routes: conditional-GAN encoders; per-image optimisation of StyleGAN codes and noise maps; pivotal tuning of the generator (under 3 minutes per face); a GAN texture prior in 3D face fitting; text or point edits on inverted faces. Trade-offs: training effort, per-image cost, fidelity against editability.ConceptGPU power cappingLimiting how much power each GPU may draw to cut the energy used per LLM output, at the cost of run time. One measurement study so far: capping A100s at 175 W running LLaMA 65B saved about 23% energy for about 7% more time on one dataset. A 150 W cap cost much more time, and the evidence covers one model, dataset and batch size.ConceptGarbled circuitsGarbled circuits let parties run a boolean circuit on wire keys that hide the values, with garblers building it and one evaluator running it. Only one card so far: a four-party ML framework that uses garbled sharing for bit-level conversions and softmax division, with the evaluator fixed to one party. Author-reported speed-ups are framework-wide, not garbled-only.ConceptGated skip connectionA layer outputs a learned, sigmoid-weighted mix of its transform and its unchanged input, so signal can pass through many layers. Gate bias starts the stack near carry behaviour. One paper so far, with author-reported results on MNIST and CIFAR-10, and no comparison to ungated shortcuts on the card.ConceptGaussian mixture modelA Gaussian mixture model (GMM) fits a density as a weighted sum of Gaussians. Five cards: WaveFake fits per-generator GMMs and scores clips by log-likelihood ratio; hypertime fits a GMM over time-projected vectors; Kuzushiji uses a mixture density network; Toto (Student-T) and LaserNet (box distributions) use other mixture forms. Author-reported.ConceptGaussian processA Gaussian process (GP) gives a mean and a variance for each prediction, and the variance grows away from training data. Ten cards use it for uncertainty, Bayesian optimization, terrain costs, track smoothing, human motion, policy gradients, dense image matching and additive kernels. All results are author-reported.ConceptGaussian splatting2DGS gives surfaces. Splatter Image: one image. MVSplat: two views. FLARE: 2 to 8 uncalibrated views. iLRM: posed views. SparseGS, FSGS: 3 to 12 photos. PhysDreamer: forces. Reduced-3DGS compresses. Hierarchical 3DGS: large scenes. GSGEN: text to 3D. SLAM: SplaTAM, GS3LAM. Gaussian Grouping, Feature 3DGS: labels. SparseGrasp grasps. Driving: DriveSplat. Forge4D: human video. GEM-Occ: occupancy.ConceptGeneralized PageRank propagationPropagates node features over several steps and mixes the results with learned or fixed weights. PageRank-style or single K-hop weights give a low-pass filter. Learned weights can go negative and fit heterophilic graphs, but are sensitive to initialisation on sparse splits. Per-node channel mixing (ACM) is a related layer. Two paper cards.ConceptGenerative adversarial network (GAN)A generator and a discriminator trained against each other, or a pretrained GAN reused without adversarial training. Twenty-nine method cards: translation, text, music, video, MRI, face editing and fitting, age progression, distillation, tabular oversampling, view synthesis. Thirteen use a GAN only as a component, benchmark, survey or metric. Numbers are author-reported.ConceptGenerative model evaluationHow to score a generative model's output beyond one frame or one sample. Covers frame-level reference metrics (PSNR, SSIM), set-level image, video and audio distances (FID, FVD, KVD, FAD), learned human-preference scores (HPS), design-oriented scores, per-task checks (MMMG) and multi-metric benchmarks (EvalCrafter). Set-level scores need a fixed sample count. All numbers are author-reported.ConceptGenerative priors for inverse problemsA generative prior is reused at inference, and a measurement or constraint enters as a gradient term. Three forms: posterior sampling by annealed Langevin dynamics for a known linear model; a point estimate under a differentiable constraint; and a diffusion-generated target fitted by a differentiable renderer, a looser match. Covers when to use which and where code departs from the paper.ConceptGlobal constraintA named rule over many variables at once, such as all-different, regular, cumulative or circuit, written as one line in a model instead of many pairwise rules. Covers the forms the PyCSP3 guide documents, how to pick one by the shape of the problem, and backend bugs to check. One source so far: a library guide with no benchmarks.ConceptGoal-conditioned reinforcement learningTrain a policy to reach a goal state given as input, with no hand-written reward. Contrastive RL learns a critic from future states the agent visits. Other variants use a goal-conditioned world model with subgoal episodes, offline landmark planning for navigation, or audio as the goal. Results are author-reported and not comparable across papers.ConceptGradient clippingRescale a stochastic gradient to a norm cap before the step, so rare very large samples cannot dominate an update. So far one card covers it: clipped inner SGD in a bilevel method for heavy-tailed noise, with a stated complexity bound and no numeric recipe.ConceptGradient conflict resolutionCombines the gradients of several losses into one update that keeps the average loss as the target while protecting the worst task. One card so far: CAGrad, which starts from the average gradient and pushes toward the worst task within a radius c. Author-reported gains over PCGrad and MGDA. Its convergence theorem and its experiments use different update rules.ConceptGradient descent stabilityFull-batch gradient descent with a fixed step size pushes the top Hessian eigenvalue up to 2/η, then keeps it hovering there while the loss still falls over long runs. Covers the Edge of Stability, momentum and SGD. Three other papers track Adam's gradient SNR, ViT gradient spikes and float32 loss spikes. Four CC BY papers, author-reported.ConceptGradient projectionRemove the part of a parameter update that conflicts with a reference gradient, so the update does not increase loss on what must be kept. Covered here: A-GEM, which uses one averaged episodic-memory constraint for weaker per-task guarantees at much lower cost than GEM, and ProGrad, which projects prompt-tuning steps against a KL term toward zero-shot CLIP.ConceptGradient-boosted decision treesTree ensembles trained by gradient boosting. XGBoost was usually the best single model and the quickest to tune in a 2021 benchmark on 11 tabular datasets. LightGBM served as the final classifier in a 2018 histology study. NODE, a 2019 differentiable tree model, reports beating tuned XGBoost on four of six datasets.ConceptGraph label propagationSpreads a few trusted values (labels or measured depths) over a neighbourhood graph so that every other node becomes a weighted average of its neighbours. Covers one variant so far: correcting a stereo point cloud with sparse LiDAR anchors, with the trade-off that anchors which disagree with the dense estimate can distort the result.ConceptGraph neural networkA graph neural network (GNN) gives each node an embedding by mixing its features with its neighbours'. Covered: forty-eight variants, from pretext-trained and contrastive encoders to lane-topology, weather, traffic and signed-graph models, plus surveys and benchmarks. The trade-off: results depend on task and graph, and the numbers are author-reported.ConceptGraph structure learningGraph structure learning lets a model decide which edges to use, or learns the adjacency from data, instead of taking the given graph as fixed. Covered: gating and pruning an observed graph with the task loss, and learning a dense adjacency from node embeddings. The trade-off: extra parameters and per-step cost, and the numbers are author-reported on different tasks.ConceptGraph transformerA graph transformer runs self-attention over a graph's nodes and adds structure: as attention biases (Graphormer), as learned positional vectors on Transformer GNNs (LSPE), as a per-layer latent graph with linear-cost attention (NodeFormer), or as a predicted pairwise graph over detected objects (D3G). Full attention grows with the square of the node count. All numbers are author-reported.ConceptGraph-based RAGAnswers questions over a whole text collection by having an LLM extract an entity graph, clustering it hierarchically, and pre-writing a summary for each community; queries run map-reduce over those summaries. Trade-off: slow, LLM-heavy indexing, and the evidence is LLM-judged on two corpora of about 1M tokens each. Vector RAG is better for single-fact lookup.ConceptGray-level co-occurrence matrixCounts how often two grey levels sit at a fixed offset in an image, then reads texture from that table with statistics such as contrast, energy and homogeneity. Variants differ in scale (pyramid or blur), in what they combine with (LBP, edges, difference images) and in how many values they produce. The one cited review runs no experiments of its own.ConceptGreedy layer-wise trainingTrains a deep model in stages, one block or layer at a time. Supervised variants give each block a shallow auxiliary classifier and freeze earlier blocks, with no end-to-end backprop (88.7% top-5 on ImageNet with 3-hidden-layer heads, author-reported). Unsupervised variants pre-train first layers with RBMs or autoencoders, then fine-tune end-to-end.ConceptGrokkingDelayed generalization: a model fits its training data, then much later learns the rule behind it. Recurrent blocks change when and whether small transformers generalize to unseen two-hop fact pairs. Also covers float32 loss spikes in long unregularised runs, which one paper argues are numerical rather than grokking. Three cards; results are synthetic or small-scale.ConceptGrouped-query attention (GQA)Query heads form G groups that each share one key and one value head, so the KV cache shrinks by H/G. G=1 is multi-query attention. A multi-head checkpoint can be converted by mean-pooling its K/V heads and uptraining briefly. On T5-XXL, GQA-8 averages 47.1 vs 47.2 for MHA at 0.28 s vs 1.51 s per sample (author-reported). Mistral 7B, Mixtral 8x7B and Gemma 2 use it in decoder-only models.ConceptGuided depth upsamplingRaising a low-resolution depth map to high resolution with a colour image as guide, using sensor depth with holes and artefacts. Covers bilinear interpolation, two classical guided baselines (JBU, FGI) and two learned networks (MSG, MSPF) benchmarked on ARKitScenes at x2, x4 and x8. Numbers are author-reported, on one dataset.ConceptHallucination detectionFlag generated text that contradicts its source or cannot be checked against it. Families: n-gram overlap, IE tuple matching, QA round-trips, NLI entailment, trained classifiers, conditional-vs-unconditional LM scoring and, for LLMs, record lookup, token-uncertainty, sample-consistency or self-evaluation checks. Trade-off: model-based checks catch more but inherit their sub-models' errors.ConceptHallucination mitigationWays to make generated text stay faithful to its source or to world knowledge: clean or augment the training data, change architecture, objective or decoding, edit activations, or check outputs after generation. For LLMs add retrieval, RLHF, model editing, debate and self-verification. Trade-off: faithfulness against fluency, diversity, cost and extra failure points.ConceptHamilton-Jacobi reachabilityA value function over the robot state that tells whether the robot can be kept out of a failure set under bounded disturbances. Its gradient can drive a safety filter. Trade-offs: ground truth comes from a grid solver, and a learned value is only calibrated on its validation distribution.ConceptHand-eye calibrationFinds the fixed transform between a camera and a robot hand from several hand motions. Compares the classical AX = XB formulation with a projection-matrix formulation MY = M'YB, and linear, closed-form and joint non-linear solvers. Author-reported: the joint solver is most accurate in simulation, and the new formulation lowers translation error on the real data sets that test it.ConceptHaptic feedback renderingTurns contact in a remote or virtual hand into a force the operator or user feels. Covers the simple force-only mapping, a variant that scales force by how far the leader and follower fingers move, physics-based contact in VR, and visual stand-ins for touch. Only two papers with cards: one human study with ten participants, one survey with no experiments. No shared metric exists.ConceptHardware-aware trainingRetraining with a simulated target chip in the forward pass, so the network tolerates device and circuit errors. In one study of 11 analog in-memory networks, 5 reached iso-accuracy after one hour of drift. Other papers use the term for a physical forward pass with a digital gradient model, latency-aware search and low-bit quantization. Author-reported; four papers so far.ConceptHeavy-tailed noiseStochastic gradient noise with only a finite p-th moment for some p in (1, 2], not finite variance, so rare very large samples are possible. Analyses that assume bounded variance do not carry over; one card gives a clipped, normalized bilevel method with complexity bounds, and no numeric recipe.ConceptHierarchical reinforcement learningA high-level policy sets a goal or subtask and a low-level policy acts under it, so decisions span several steps. Eleven cards so far, from text-GAN managers and robot subtask planners to humanoid, bipedal, navigation, dexterous and offline-maze control.ConceptHierarchical vision transformersVision transformers that shrink the token grid in stages and widen the channels, as CNNs do. The variants differ in how they reduce the grid or the attention cost: PiT and P2T pool, EdgeViT and LITv2 mix local and global attention, HiFormer and SwinNet build on Swin pyramids. Trade-off: gains depend on recipe, hardware and task, and the numbers are author-reported.ConceptHindsight relabelingTurning failed or suboptimal rollouts into training samples that show how to recover. Hindsight Planner rewrites the task text from structured arguments and has an LLM reflect on the failure with the expert rollout in view. Author-reported: removing it lowers ALFRED valid SR by 2.08 (seen) and 2.68 (unseen). Only one card so far.ConceptHomomorphic encryptionComputation on encrypted data, so the party holding the data never sees it in clear. Only one card so far: a ResNet-20 run on encrypted CIFAR-10 images with the CKKS scheme and bootstrapping. Author-reported: about 98.67% agreement with the plaintext network on 75 images, at about 4 hours per image.ConceptHuman evaluationJudging generated outputs, dialogues or robot behaviour with people and turning the judgments into a verdict. Averaged Likert ratings can reverse the true preference when systems are close; a per-annotator probability question recovered more expected orderings in one study. Pairwise trials feed a ranking model. Results are author-reported, from twenty-one papers.ConceptHuman-AI collaborationSplitting a task between a person and an AI system, each doing the part it suits, with both improving over time. Covers the Hybrid Intelligence framing, a survey of collaboration patterns and build-time guidance, an AI-drafted writing workflow and a self-report scale for checking AI output. All evidence is author-reported; no card tests whether the combination beats either part alone.ConceptHuman-in-the-loopA person stays inside an AI system's decision or learning loop: choosing outputs, teaching through labels or corrections, sharing control, or limiting autonomy. Twelve cards cover these. Evidence ranges from position papers and a survey to user studies, trials and robot experiments.ConceptHypernetworkA network whose output is the weights of another network. Two variants here: a server-side hypernetwork maps a learned client embedding to that client's weights, and only weight changes travel back (federated); a document hypernetwork writes a rank-8 LoRA adapter from a document's activations in one forward pass. Both trade up-front training for fast, per-input weights.ConceptHyperparameter optimizationAutomated search for settings fixed before training, such as learning rate, scored by each trial's validation result. Options run from grid and random search through successive halving and Bayesian optimization to hypergradients. The choice depends on the trial budget and on whether partial runs predict final results. Tuned scores are noisy, so compare methods over repeated runs.ConceptImage captioningFour ways to generate captions: a chest x-ray report decoder with co-attention over regions and tags; a transformer captioner pre-trained on relaxed web alt-text; BLIP's captioner that relabels noisy web images; and ClipCap, which maps a CLIP embedding to a GPT-2 prefix. All numbers are author-reported and not comparable across papers.ConceptImage copy detectionImage copy detection checks whether a query image contains an object that also appears in a reference set, by retrieving the nearest reference images with a pretrained encoder. Variants differ in the encoder and in how patch or token features are compared. Only one card so far, on diffusion-model replication, with author-reported results.ConceptImage deblurringRecover a sharp image from one blurry photo. Coarse-to-fine stacks are accurate but costly. MIMO-UNet puts the multi-scale idea in one U-Net, with a small gap to top GoPro results. DiffIR diffuses a 256-value prior in 4 steps and tops its GoPro table, after two-stage training.ConceptImage denoisingImage denoising removes noise from a picture: Gaussian noise at a known level, real sensor noise, or Poisson photon noise. Five cards: MambaIR and GRL disagree on which Gaussian model is best; DGUNet is within 0.1 dB of MambaIR on SIDD and DND; RIDNet, a one-pass blind denoiser, scores 39.23 dB on DnD; WIRE fits a wavelet implicit network to one noisy image, with no training set.ConceptImage enhancementRestoring the light, exposure or colour of dark, backlit or underwater photos. Four variants: a tiny supervised transformer for light and exposure, an unsupervised CLIP-guided illumination map for backlit photos, and two paired underwater models, a transformer GAN and a CNN that blends RGB and HSV. The trade-off is paired data and model size against unpaired data and CLIP compute.ConceptImage fusionCombining images or sensors of one scene into one result. Three families: rule- or wavelet-based fusion of registered medical scans; a second sensor as a transformer token (hyperspectral plus LiDAR); and camera and LiDAR feature maps fused in a shared bird's-eye-view grid. The trade-off is interpretability and data needs against accuracy and compute cost.ConceptImage manipulation detectionFind edited regions in an image, or decide if a face or whole image is fake. Four cards: RGB-N, a two-stream box detector (NIST16 pixel F1 0.722); PCL, face-video patch consistency (no official code); FDFL, a face-video detector with a single-center loss; CIFAKE, a whole-image classifier. All results are author-reported.ConceptImage quality assessmentScoring how good an image looks, either against a reference image (full-reference) or without one (no-reference). Two no-reference cards: a metric trained on human ratings of super-resolution outputs, and LIQE, a CLIP-based model trained on six quality datasets. Trade-off: the cards are validated on different data, so their scores are not comparable.ConceptImage registrationImage registration aligns images or views so corresponding points coincide. A 2013 fusion survey names techniques without an algorithm. A 2018 3D CNN learns prostate MR-to-ultrasound warps from sparse labels. BARF (2021) aligns NeRF camera poses photometrically, coarse to fine. Trade-off: each needs labels or RGB views of a rigid scene, and reports narrow tests.ConceptImage super-resolutionLearned upscaling of a low-resolution image or field, and ways to judge or explain the result. Covers ESRT, GRL, MambaIR, StableSR, SeeSR, DiffIR, eDiff-I, BasicVSR, RBPN, ACNN-SR, WIRE, a physics-informed diffusion model for flow fields, KernelGAN blur-kernel estimation, LAM attribution and a no-reference metric. Protocols differ.ConceptImage thresholdingSplitting an image into classes by comparing pixel values with cut points. Multilevel thresholding picks several grey-level cuts by an Otsu or Kapur score, found by metaheuristic search (MAAPO). Single-threshold classification cuts one colour index, such as Excess Blue, at a value tuned on labelled pixels. Two sources, one method each.ConceptImage watermarkingMakes a generator's images carry a hidden key so a detector can later flag them or name the user who made them. One approach, in-generation watermarking, fine-tunes the decoder of a latent diffusion model so no separate step is needed. Author-reported: 0.99 bit accuracy on unmodified output, 0.92 after a combined crop, brightness and JPEG edit.ConceptImage-to-3D generationTurning one photo into a 3D model, by generating views or by optimising a 3D field with a 2D diffusion prior. Five variants: One-2-3-45++ (objects, about 60 s, author-reported), PSHuman (full-body people, SMPL-X), SV3D (orbit video, non-commercial weights), Make-It-3D and RealFusion (per-object optimisation, tens of minutes to hours).ConceptImage-to-image translationMaps images from one domain to another while keeping the input's content. Covers ten variants: unpaired and paired GANs, one multi-domain GAN with style codes, training-free diffusion editing, flow-based style transfer, single-pair ViT-feature generation, and translation used for segmentation adaptation or anomaly detection. The choice depends on paired data, labels, and model access.ConceptImitation learningLearning a policy from recorded behaviour instead of a hand-written reward: behaviour cloning, nearest-neighbour lookup, return-conditioned sequence models, imitation terms in offline RL, vision-language-action and tactile policies, planning from logged video, imitating solver branching, expert hints for LLM agents and style imitation, with the trade-offs in data, compute and evaluation.ConceptImpedance controlControl that makes a robot act like a spring and damper toward its environment, so contact force follows from stiffness and damping. Four cards: task-space reaching with a contact test, a rehabilitation teleoperation pair, co-manipulation with a stiffness hand-over, and joint-space impedance under a learned policy. Evidence is author-reported, mostly qualitative.ConceptImplicit bias of gradient descentWhich solution gradient descent selects when many fit the data. For linear models on separable data, the parameterization sets the limit: fully connected layers give the max-margin direction at any depth, convolutional layers favour sparse Fourier spectra, more so with depth. Mostly theory. For nonlinear networks, full-batch descent settles at a sharpness set by the step size (author-reported).ConceptImplicit neural representation (INR)An implicit neural representation is an MLP that maps coordinates to a signal, such as colour, occupancy or signed distance, so the signal lives in the weights. Four choices matter: the nonlinearity (WIRE's Gabor wavelet), the supervision (iSDF, online from depth), a latent code shared across objects (NeuSE, Neural Scene Graphs) and time as the coordinate (StyleGAN-V).ConceptImportance weightingReweighting samples so their average estimates a quantity under another distribution, or so a sample stands in for the items it represents. Covers density-ratio validation under covariate shift, inverse-inclusion Shapley weights, off-policy RL ratios and replay, defensive mixture sampling for policy gradients, and heuristic weights for VLM fine-tuning. Author-reported results only.ConceptImprecise probabilityReplaces one probability per label with a family of compatible distributions (a credal set) built from split-CP p-values. Its spread gives uncertainty summaries that use the whole score profile, computed in linear time after sorting. Author-reported gains with LAC scores; probability baselines win under APS, RAPS and SAPS. One paper so far.ConceptIn-context learningPut k worked input-output examples, or a plain task description, in the prompt and let a frozen model continue. Random labels cost only 0-5 points for most classifiers, but retrieving similar examples added about 17 F1 for GPT-NER on CoNLL2003. Results also move with prompt layout, example choice and the scorer used.ConceptIn-hand manipulationChanging an object's pose while a gripper, palm or finger holds or lifts it, without putting it down. Four mechanisms: tilting a large object against supports, a suction palm with belts, tape-spring arms that roll and convey, and vibration on a parallel gripper. Author-reported, on different setups.ConceptInductive biasThe built-in assumptions a learner uses to prefer one explanation over another, which every generalization claim depends on. A position paper argues that deep learning needs extra biases from high-level, verbalisable causal structure for out-of-distribution and systematic generalization. It proposes nine biases, gives no algorithm and no experiments (author-reported design arguments).ConceptInfluence functionsInfluence functions estimate how a trained model changes when training rows change, using one Newton-style step instead of retraining. Two variants: certified removal of rows for L2-regularised convex models, with random loss perturbation (250x to 870x cheaper than retraining, author-reported); and closed-form replacement of features or labels, certified only for strongly convex losses.ConceptInformation bottleneckInformation bottleneck (IB) appears in two compiled papers. One reads deep-network training as fitting, diffusion and equilibrium phases, with layer compression that the paper finds mostly saturation. The other uses a graph IB theorem to motivate a learned edge-drop augmentation for self-supervised graph contrastive learning.ConceptInformed machine learningTraining a model on data plus explicit prior knowledge, such as equations, logic rules, simulator output, invariances or human feedback. The knowledge can enter through the training data, the hypothesis set, the learning algorithm or the final hypothesis. One survey card so far, with no experiments of its own; it gives a taxonomy, not measured results.ConceptInput convex neural network (ICNN)A feed-forward network whose output is convex in its input, built by keeping hidden-to-hidden weights non-negative and using convex, non-decreasing activations. One card so far: used as the two potentials of a minimax, quadratic-cost optimal transport map learned from samples (author-reported, 784-D Gaussian pairs). Single variant; no comparison yet.ConceptInstance segmentationGive each object, map element or 3D scan object its own mask. Copy-Paste pastes masks at no inference cost; BCNet adds an occluder-aware mask head; Mask3D predicts 3D scan masks with Transformer queries; HDMapNet clusters bird's-eye-view embeddings; VisTR predicts per-clip video masks; EQL v2 and DisAlign rebalance rare classes; connected components split buildings. Most need labels.ConceptInstruction tuningFine-tuning a pretrained model on instruction-style pairs so it follows requests. Forty-six carded papers vary the data and method: curated pairs without RLHF (LIMA), prompted task mixtures (T0, Flan), model-generated or crowd-written dialogue (Self-Instruct, CAMEL, OpenAssistant), domain and LoRA tuning, and image, video, speech and driving inputs. Results are author-run.ConceptInter-annotator agreementAgreement between annotators, or between a model and experts, shows whether labels are reliable enough to train or judge on. Hate-speech studies report Krippendorff's alpha of .18 to .38, and 0.815 for round-2 originals in a dynamic collection. A language model coding against expert codebooks reaches Cohen's kappa of .61 and .38, against .88 and .90 for experts. These measures are not comparable.ConceptInteractive segmentationClicks or boxes refine a mask. Zero-shot SAM with boxes is the simplest option for 2D medical images; a promptable 3D model suits volumes, where per-slice prompting is costly. A diffusion model segments one object crop with no clicks. Results are author-reported and simulated from ground truth.ConceptInterior-point methodSolve a convex QP by keeping iterates strictly inside the inequality constraints through a log barrier. For long-horizon sparse MPC, a structure-exploiting interior-point solver (HPIPM) is the review's recommendation. Only one review card supports this page, and its benchmark claims are relayed from other work.ConceptIntermittent computingRunning programs or neural networks on devices that lose power repeatedly, such as harvested-energy sensors. Each power-on period has a fixed energy budget, so progress must be saved and resumed after each outage. One paper so far, author-reported on a 2 KB microcontroller. Trade-off: a smaller network and extra saving of progress, with latency and accuracy that depend on the energy budget.ConceptInternal learningTraining a model on the single input it will process, using augmented copies or patches of that input, not a dataset. Covered: a CLIP-guided editor, a CLIP-guided stylizer and an internal GAN for blur kernels. Trade-offs: training cost for every input, no reuse across inputs, and dependence on pretrained models in the CLIP variants.ConceptInterpretable machine learningModels whose structure a person can read: sparse trees, scoring systems, neural and additive Gaussian processes, prototype networks and few-feature linear models. A survey urges trying them first in high-stakes decisions. A tabular study finds neural additive models trail the best black box by a small margin, and SHAP and LIME rankings move with the model. Results are author-reported.ConceptIntrinsic dimensionMeasures how many free parameters a fine-tuning task needs by training only d numbers pushed through a fixed random projection onto frozen pretrained weights. The smallest d reaching 90% of full fine-tuning is the task's intrinsic dimension; for RoBERTa-Large on MRPC it is 207 to 322 numbers. Values are upper bounds and depend on the metric and the projection.ConceptIntrinsic motivationRewards an agent for its own prediction surprise or uncertainty instead of a task reward, so it can explore before any goal exists. Variants: ensemble disagreement, forward-model error with an inverse model, and error against a fixed random target. One card: in a CARLA study, disagreement led the other two zero-shot on an unseen town; author-reported, one simulator.ConceptIntrusion detectionFlags attack traffic: network floods (DoS/DDoS) and SQL injection in web requests. Two supervised methods, each author-reported on one setup: a small sigmoid network on IoT traffic (99.4% accuracy), and learned weights over ModSecurity rules (99.02–99.45% TPR at 1% FPR, paranoia level 4, vs 68.55% for stock rules). Also a labelled NetFlow dataset rebuild; it trains no detector.ConceptInvariant learningTrain a model to rely only on features whose link to the label holds across environments. Covered: EERM, which edits one graph into virtual environments when labels are missing; IRM, as a survey summarises it; and Fish, which matches gradients across labelled training domains. Trade-off: EERM's editors are dense n × n matrices, costly on large graphs.ConceptInverse dynamics controlUses the robot's dynamics model to map a commanded acceleration to joint torques. Two cards: a quadruped controller that enforces the exact friction cone and splits joint tracking into a null space, and a GPU planner that evaluates inverse dynamics inside trajectory optimisation to respect torque limits. Both results are author-reported.ConceptInverse kinematics (IK)Find joint values or velocities that reach a target pose. Five variants: QP velocity IK with soft collision barriers, Jacobian IK for parallel limbs, nested Newton IK for closed loops, learned-collision position IK for legged robots, and GPU Levenberg-Marquardt IK for high-DoF humanoids. Results are author-reported; only cuRoboV2 has official code.ConceptInverse reinforcement learningInferring a reward function from demonstrations. Covered: a maximum-likelihood observer that assumes the demonstrator is doing the task, a pedagogical variant that models a teacher who shows it, cooperative IRL, and adversarial (discriminator-based) IRL. Behaviour alone under-determines the reward. Results are author-reported, mostly on small or lab-collected data.ConceptIsolation forestScores points as anomalous when random cuts isolate them after few steps. Covers the classic axis-aligned forest, its hyperplane extension, and Deep Isolation Forest, which cuts random neural representations instead of raw features. Results are author-reported on tabular, graph and time-series sets.ConceptIterative closest point (ICP)Iterative closest point aligns two point clouds by alternating correspondences with a closed-form rigid fit (SVD). Variants differ in correspondences, residuals and weights. Classic ICP assumes equal point counts, which partial views break. Learned matching (PRNet) and semantic weighting (SuMa++) respond; SoGraB uses ICP only to pre-align.ConceptKV-cache managementHow to lay out, allocate, share and bound the attention key/value cache during generation. Paging it into fixed-size blocks behind per-sequence block tables removes most fragmentation and enables copy-on-write sharing, at the cost of slower kernels. Sink-plus-window and eviction schemes cap stream caches by dropping context. Grouped-query attention shrinks it per token but changes the model.ConceptKernel matrix reductionReduce a pairwise kernel or distance formula over two point sets on the GPU without storing the full matrix, with autodiff through the formula. KeOps is the one paper so far: author-reported, a Gaussian product over 1M 3D points takes 1.39 s where PyTorch runs out of memory from 100k. Trade-off: a compile per new formula, and evidence from one GPU and one kernel.ConceptKeyphrase extractionPicking the few words or short phrases that describe a document. Covered so far by one card: Bi-LSTM and attention taggers for tweets, with per-word cognitive features (EEG and eye tracking) added. Author-reported F1 gains over a plain Bi-LSTM baseline, which also differs in architecture; the released code differs from the paper and does not reproduce its best setting.ConceptKeyword spottingDetecting a few target words in audio that is mostly silence or noise, on-device under tight compute and energy limits. Covers the Speech Commands benchmark, its Top-One and streaming protocols, building a similar corpus, and a spike-coded version of the same audio. Two cards; no detector comparison yet.ConceptKinematic bicycle modelA force-free vehicle model with no lateral slip, where each axle is one wheel at its centre and every block moves along its own heading. Two papers use it: an articulated tractor-trailer tracked by nonlinear MPC, and closed-form arc primitives for a hybrid A* search in a car-like planner. Results are author-reported, from simulation and one real vehicle.ConceptKnowledge base question answeringKnowledge base question answering (KBQA) answers a question with knowledge-graph entities or a logical form. Temporal KBQA adds when facts held: CronKGQA reaches author-reported Hits@1 0.647 vs 0.288 for the best baseline. GrailQA's BERT ranker reaches 58.0 F1 overall. KV Retrieval Net writes KB values into dialogue replies (entity F1 48.0). CronKGQA is weak on multi-fact questions.ConceptKnowledge distillationA student is trained to match a teacher's outputs, not only labels. Targets can be class probabilities, stored top-K probabilities, feature maps, pairwise relations, EMA-teacher patch tokens, density at 3D points, few-step sampler outputs or written rationales. Trade-off: a teacher pass or stored targets, often an extra training run, and gains that depend on teacher and benchmark.ConceptKnowledge graph completionPredicts missing facts in a knowledge graph by scoring candidate triples, then adds the high-scoring ones. Embedding models are the common choice; their evidence is a 2023 survey's table that mixes datasets and metrics. GraIL scores from the local subgraph with no entity embeddings, so it handles unseen entities (author-reported wins on 12 inductive splits). Trade-off: a subgraph per candidate.ConceptKnowledge graph embeddingKnowledge graph embedding maps entities and relations to dense vectors and scores each triple for plausibility, so missing facts can be predicted. Families: tensor factorization, translation (TransE), neural scoring (SME, ConvKB, R-GCN), region-based and Gaussian embeddings, and temporal (TComplEx). Trade-off: static numbers come from one mixed-protocol table, and training details are not stated.ConceptKnowledge graph reasoningKnowledge graph reasoning infers a relation between two unlinked entities. GraIL scores unseen entities from subgraph shape (author-reported best inductive AUC-PR on 12 splits). CronKGQA answers when-questions over timed facts (author-reported Hits@1 0.647, but 0.392 on complex questions). A 2020 multi-hop QA baseline scores 5.41 joint F1, so pick a method by task.ConceptKnowledge injectionKnowledge injection adds facts a pretrained LLM lacks, either by continued training on documents or by retrieving them at query time. On 7B models, Ovadia et al. report retrieval beating fine-tuning in almost all settings (author-reported); paraphrased training text helps fine-tuning. Trade-off: retrieval adds an index and per-query tokens, fine-tuning adds training cost and may forget.ConceptLLM agentA language model that loops through reasoning, tool calls and observed results toward a goal. Forty-nine reuse-licensed papers cover single agents, planners, agent memory, trained agents, multi-agent teams, self-improving harnesses, agent evaluation, failure attribution and surveys. Most results are author-reported on different tasks, so no design is shown to win in general.ConceptLLM ethicsEthical and social risks of language models: risk taxonomies and a 1,725-risk database, auditing blueprints, bias, toxicity and stereotype measures, cultural and religious bias, memorised training text, web corpus filtering, carbon cost, authorship and copyright, education, mental-health and clinical use, and agency. Most results are author-reported; cards note how each was measured.ConceptLLM evaluationHow to measure what a large language model can do, and what changes the reported number: prompt, format, decoding, model snapshot, metric, judge and ground truth. Variants cover fixed-prompt benchmarks, setting and prompt sweeps, complexity curves, minimal-edit probes, LLM and human judges, simulated participants, reading-time fit, bias probes, safety rates, surveys and hallucination checks.ConceptLLM planningLLM+P writes PDDL for a classical planner. Text2Motion checks LLM skill sequences with learned feasibility models. KnowNo asks a human when calibrated options are ambiguous. Chameleon, LLM-Planner and Hindsight Planner pick steps from examples, the last two replanning. DCIP edits a grid map; DexDiff uses a fine-tuned VLM plan. Each needs a model, data, tools or demonstrations.ConceptLLM self-evaluationA language model judges its own output, or a partial one, and the judgement sets the next step: a rewrite, a re-run, a retry with a written lesson, a search choice or a rank. Mostly no training, but every judgement is an extra model call and is only as good as the judge. Author-reported gains are largest where the judge sees concrete errors, such as failing tests, and small on math.ConceptLLM servingSystem design for serving many concurrent LLM generation requests at high throughput: iteration-level batching, request scheduling, preemption with swap or recompute, and multi-GPU execution. Throughput is largely set by how many requests fit in a batch, so serving engines are built around KV-cache memory; gains appear mainly when serving is memory-bound.ConceptLLM simulated participantsLanguage models as stand-ins for survey respondents. One 2023 study finds that blinded prompts confound a simulated price effect with invented customer details. Stating the randomisation design (unblinding) lowers error against a human survey. Adding covariates to the prompt gives non-monotone error and focalism. Trade-off: prompt-only cost versus fine-tuning on survey data.ConceptLLM tree searchSearch over a language model's own intermediate reasoning steps: the model proposes candidate next steps, scores partial solutions, and breadth-first or depth-first search keeps promising branches and backtracks. No training. Tree of Thoughts reports 74% vs 4% for CoT on Game of 24 (GPT-4), at $0.74 vs $0.47 per problem for best-of-100 CoT. LLM+P instead hands the search to a classical planner.ConceptLLM-as-a-judgeUsing a language model to grade other models' outputs in place of human raters. Human agreement runs from 50.8% (forced choice against a gold statement) to 78% (pairwise preference). A zero-shot GPT-4 score reached 89.8% system-level accuracy on machine translation (author-reported). The result depends on the task and the judge: T0 did not reproduce the teacher ordering that text-davinci-003 did.ConceptLabel assignmentDecides which predictions count as positives (trained to match a ground-truth object) and which as negatives. One-to-one assignment gives each object one prediction: no duplicates, slower learning. One-to-many gives several: faster learning, but duplicates to remove at inference. Hybrid schemes train with both and keep one-to-one at inference. Scoring by receptive field helps tiny objects.ConceptLabel smoothingSoftening one-hot training targets so the classifier is less sure of its training labels. Covers a plain smoothed cross-entropy, smoothing carried by mixup soft targets, and TSLA's smooth-then-hard schedule. Trade-offs: the number means different things by convention, and the cited gains come from larger training recipes, not from label smoothing alone.ConceptLangevin dynamicsSamples from a distribution given by an energy, by repeating a gradient step downhill plus Gaussian noise. Only one card so far: COLD decoding applies it to a soft text sequence with weighted constraint terms, then maps each position to a token. Author-reported: 94.50% keyword coverage on CommonGen, at higher perplexity than NeuroLogic.ConceptLanguage identificationAssigning a language label to text lines, documents or speech, used to sort web-crawled corpora and to benchmark speech models. Covers line-level and document-level labelling, n-gram and Transformer LangID filters, and speech LangID. Every variant guesses, and the audited corpora show that wrong labels remain common, so check labels before use.ConceptLanguage model pretrainingTrain a Transformer on unlabeled data with a self-supervised objective, then reuse it by fine-tuning, prompting, freezing or as a component. Covers XLNet, PaLM, Qwen, French, Dutch, multilingual, byte-level, materials, chemistry, dialogue and knowledge-graph models, frozen transfer, word budgets and corpus effects. Trade-off: objective and data versus compute and use.ConceptLanguage model surprisalSurprisal is the negative log probability a language model gives a token in its context. Word surprisal predicts human reading times, scored as held-out ΔLogLik over a baseline. Sequence perplexity of a prompt ranks zero-shot prompts without labels. Both uses score different outcomes, so their numbers are not comparable. Define tokenisation and the spillover window before comparing.ConceptLarge reasoning modelsLanguage models that write a long reasoning trace before answering, built in rather than prompted. Thinking models lead only at medium puzzle difficulty, and broad task results rest on small, author-graded samples. A 1.5B model trained with RL reached 46.7% on AIME24 (author-reported) for about \$42. Pruning a student's own long traces cut 7B GSM8K tokens by 64% (author-reported).ConceptLarge-batch trainingHow to train with a large global batch across many workers without losing test accuracy. Options: linear learning-rate scaling with warm-up, a batch-size ramp, local SGD, post-local SGD (local steps after the first learning-rate decay) and hierarchical local SGD. Author-reported on CIFAR and ImageNet. Trade-off: fewer syncs against tuning effort and a first phase that still syncs every step.ConceptLasso regressionLasso regression is an L1-penalised linear model that sets some coefficients to exactly zero. Safe screening rules (DPP, with EDPP the tightest) discard features that are provably zero before each penalty on a grid is solved. A binomial Lasso that selects text features for AI-versus-human classification is a separate, small-sample use. Speed-ups are author-reported.ConceptLatent diffusionDiffusion on a learned, smaller code: an autoencoder compresses samples, a denoiser works on codes, and a decoder maps them back. 30 cards cover motion, 3D, editing, video, animation, driving, restoration, depth, safety, watermarks and robot control. Latent size, condition entry and guidance matter most. Results are author-reported, not comparable.ConceptLatent-space editingEdits an image by moving its code in a pretrained generator's latent space, then decoding. Ten cards cover trained encoders, flows, CLIP-guided optimisation, per-photo tuning, noise-map embedding, face-age autoencoders, StyleCLIP, point dragging and motion. Trade-offs: labels and training versus minutes per image. Author-reported numbers.ConceptLayout-to-image generationMakes a diffusion model place objects where you ask, given as boxes with phrases, boxes with classes, painted or region masks, or a layout an LLM writes. GLIGEN and LayoutDiffusion train box-conditioned layers; the mask-based controls, box guidance on attention and LayoutGPT's planning need no training. Covers inputs, training cost and the constraints that choose between them.ConceptLearnable activation functionsLearnable activation functions put a trainable univariate function on every edge instead of a fixed nonlinearity on every node. One card so far, KAN, which uses spline edges and reports small models that match larger MLPs on small function-fitting and PDE tasks. Results are author-reported, on small problems, and trained with LBFGS, so check the constraints before adopting it.ConceptLearnable delaysGives each recurrent connection of a spiking network a trainable travel time, learned with its weight by surrogate gradients. The delay is spread over neighbouring steps during training, then rounded to integers for inference. Axonal (per neuron) and synaptic (per connection) variants; the reported gains over fixed delays are small and not significant in every case.ConceptLearned simulatorA learned simulator predicts the next state, or a short plan, and is rolled forward by feeding its output back in. GraphCast forecasts global weather with a mesh graph network. TrafficSim simulates multi-agent traffic from driving logs and reports 0.50% scenario collisions. A survey covers planning with one-step models but reports no results.ConceptLearned sparse retrievalScores a query against a document with learned, non-negative weights on vocabulary terms, so only shared terms count. One card so far: M3-Embedding's sparse head, trained with its dense and multi-vector heads. Strong on long and English documents in the paper, weak across languages, and it needs self-distillation to work well.ConceptLearning from demonstrationTeaching a policy a task from recorded examples instead of hand-coding it. Cards cover planning, behaviour cloning, nearest-neighbour lookup, reward inference, LLM prompting and teleoperation kits. Evidence is author-reported, mostly small or simulated, plus one 100-task real-robot study.ConceptLearning with noisy labelsWeb-scraped, crowd-tagged or production labels are often partly wrong. Options: clean before training (score filtering, relabelling model disagreements, or audited cross-split checks), isolate noise in the loss (sub-center ArcFace) then drop and retrain, reweight or select during training, or model uncertain labels as a third state. Trade-off: annotation time versus automatic, model-based cleanup.ConceptLearning with rejectionModels that may decline to answer and pass the input on, usually to a person. Covers separated, dependent and integrated rejectors, how to set the rejection threshold, and how to combine rejectors. One survey card; it runs no experiments, so it gives no tuned numbers.ConceptLearning-rate schedulesHow to set the step size over a training run. Theory for one-pass SGD with a known horizon: decay from step 0 on easy tasks, or hold the stability limit and decay at the end on hard tasks. Practice: warm-up then cosine or one-cycle, step decay with a batch-scaled peak, flat-then-cosine, or halving on validation plateau. Constants are often not stated, so the peak must be tuned.ConceptLearning-to-rank losses over candidate setsRanking-style losses over a set of candidate outputs. Decision-focused learning: the true optimum is compared with a cache of feasible solutions, so no solver call is needed in the backward pass (four variants). Summarization: BRIO orders generated summaries by ROUGE with a margin loss. Not search-ranking LTR.ConceptLeave-one-out analysisRemove one row, entry or training record, recompute, and compare with the full result. In spectral analysis it is a proof technique for entrywise error bounds, which the monograph reports as much smaller than ℓ2 bounds. In membership inference it is the ideal reference: a model retrained without the record, which needs the other n−1 records, so attacks approximate it.ConceptLevel-of-detail renderingShow far or small content at lower detail so each frame stays affordable. Two cards: a Gaussian-splatting hierarchy with merged parents and blended level switches, and a NeRF with one output head per scale. Results are author-reported on different scenes. Chunked training helps very large captures.ConceptLiDAR-inertial odometryLiDAR-inertial odometry estimates a moving platform's pose by fusing laser scans with IMU readings. The IMU predicts motion between scans, and each scan corrects the prediction. Three papers show two designs: iterated error-state Kalman filters (R2LIVE, R3LIVE) and an iSAM2 factor graph with scan matching (LVI-SAM). Accuracy figures are author-reported.ConceptLicense plate recognitionReading the characters of a vehicle plate from camera images or video: a chain of detectors (vehicle, plate, characters), then separate letter and digit recognisers, with a per-vehicle vote over frames. One design so far, built for Brazilian plates, reported at 93.53% on SSIG and 78.33% on UFPR-ALPR with voting. Choose by plate layout, whether you have video, and latency.ConceptLikelihood-based text evaluationScoring text or answers by how likely a pre-trained language model finds them, with no human labels. BARTScore averages token log-likelihoods given a reference. Domain-conditional PMI subtracts a domain-prompt likelihood to offset answers with many wordings. LangFlow reports generative perplexity. Theory-of-mind probes pick the highest conditional likelihood.ConceptLinear attentionLinear attention replaces softmax attention so cost grows linearly with length. Variants differ in the kernel (random features, elu-plus-one, Nyström landmarks, channel cross-covariance, Gumbel-noised features), state decay, gated matrix memory (mLSTM) and chunking (FLASH). Short inputs gain little; RFA-trained models did not transfer to softmax without finetuning.ConceptLinear probingFitting a linear classifier on frozen features, with the network never updated. In few-shot use, the probe is refit per test episode from labelled support images. A supervised ResNet-12 with an L2-normalised logistic-regression probe reaches 62.02% (1-shot) and 79.64% (5-shot) on miniImageNet 5-way, author-reported. Trade-off: cheap to fit, but capped by how good the frozen features already are.ConceptLinear-quadratic regulator (LQR)Optimal linear state feedback for a linear system with a quadratic cost, found by solving a Riccati equation. Two cards use it in legged and humanoid control: a time-varying LQR gives a value-function surrogate for a walking QP, and a Riccati recursion computed by a parallel scan solves the inner problem of a GPU MPC. All results are author-reported.ConceptLipschitz constraintCapping each layer's operator norm (its worst-case stretch under l1, l2 or l-inf) and projecting the weights back under the cap after every update, so the network's Lipschitz bound is the product of the caps. Trade-offs: only l1 and l-inf caps are exact, the l2 cap is an estimate, and the cap value is sensitive.ConceptLocal SGDLocal SGD lets each worker take H steps between model averages, trading communication for drift. Post-local SGD runs mini-batch SGD until the first learning-rate decay, then switches to local steps. Federated variants add a global gradient correction to each local step. Reported gains depend on the setting and the baseline.ConceptLocal binary patternsEncodes each pixel as a binary code by comparing it with neighbours on a circle, then describes a texture by those codes. Variants add rotation invariance (minimum over circular shifts), uniform-pattern labels that shrink the histogram, or combination with co-occurrence and edge features. The one cited source is a 2019 survey with no experiments of its own, so no LBP-only accuracy is given.ConceptLocal feature matchingFinding points in two inputs that show the same scene point. Detect-and-describe pipelines, or dense matching without a detector (LoFTR, RoMa). In 3D, RoPS describes mesh patches, PREDATOR and PRNet pick points when scans overlap little, and Deep Global Registration and PointDSC filter matches. FeudalNav uses SuperGlue matches to steer navigation.ConceptLocal reference frameA coordinate frame attached to a point of a surface, built from the geometry near that point, so that local data can be expressed the same way under any pose. Two designs are covered: a scatter-matrix frame with a sign rule that gives unique axes and one-correspondence pose, and a normal-based frame that places grasp candidates on point clouds. All results are author-reported.ConceptLocal surface descriptorsA local surface descriptor encodes the 3D shape of a small patch around a keypoint so the same patch matches across noise, mesh resolution changes, occlusion and clutter. One card so far: RoPS, a fixed-scale descriptor of 135 values built on a sign-disambiguated local frame. Its author-reported recognition rates are 100% on Bologna and 98.9% on UWA.ConceptLocal-global attentionLocal-global attention pairs a nearby window with a route to distant text. Gemma 2 alternates local and global layers. LongT5 adds one summary token per 16-token block in every layer. HiLo splits each layer's heads between windows and pooled keys. All results are author-reported, and no paper here compares against another's variant.ConceptLong short-term memory (LSTM)A gated recurrent cell with an additive cell-state update. xLSTM adds exponential gating and matrix memory, and reports lower SlimPajama perplexity than Llama and Mamba at 15B tokens. LSTMs also classify skeletons and tactile series, decode dialogue, correct vehicle forces, decode EMG, detect anomalies and run PPO locomotion policies. Results are author-reported.ConceptLong-tailed recognitionRaising accuracy on rare classes when a few classes dominate training. Covers Copy-Paste with two-stage class-balanced retraining, equalization loss v2 for end-to-end detection, weight balancing and distribution alignment for classifiers, rare-predicate methods for scene graphs, and rarer-word web data. Numbers are author-reported; compare within one recipe.ConceptLongitudinal data analysis with tree ensemblesAdapting random forests to repeated measurements of the same subjects, whose rows are correlated. Three routes: resample whole subjects, add mixed-effects random effects, or treat each subject's trajectory as one multivariate response. Trade-off: simple sampling fixes ignore time and random effects, while mixed-effects hybrids are richer but harder to fit and have little high-dimensional evidence.ConceptLoop tilingCutting a layer's loop nest into tiles so each piece fits in an accelerator's small on-chip memories, then overlapping fetches with compute and unrolling inner loops for throughput. The trade-off is on-chip buffer size against off-chip traffic and resource use, and tile sizes must be picked per layer. The only source is one FPGA survey that quotes numbers from other works.ConceptLooped transformerA transformer that applies one shared block of layers R times in a row, so effective depth is layers times R without new weights. Reported gains: unseen fact combinations and chains longer than training, both via more loops, with overthinking beyond a point. Variants: fixed or sampled R, and stochastic latent guidance with majority voting. All evidence comes from small synthetic or puzzle tasks.ConceptLoss disentanglementSplits one box regression loss into one term per parameter group, so each term sees only its own group from the prediction. The paper reports more stable training for monocular 3D detection. Its ablation gains from this step alone are small, and the cost is k times the loss terms.ConceptLow-precision arithmeticStore or multiply numbers in fewer bits than FP32 to cut memory and raise throughput. Twelve wikillm papers cover it: two surveys, FP8 attention, NVFP4 and INT4 LLMs, 4-bit clipping, 16-bit storage, FP32 decoding, FP16 kernels, float32 rounding that causes loss spikes, and bfloat16 versus float32 on TPUs.ConceptLow-rank adaptation (LoRA and variants)Fine-tune a frozen pre-trained model by training a small low-rank product L1 L2 next to each chosen weight, so only the adapter gets gradients and optimizer state. QLoRA stores the frozen base in 4-bit NF4 to fit 65B fine-tuning on one 48GB GPU, and finds adapters must cover all linear layers. Trade-off: the full base still has to sit in memory during training.ConceptLyapunov optimizationLyapunov drift-plus-penalty turns a long-run cost minimisation under queue stability into one decision per time slot. Each slot minimises a weighted penalty plus the drift of a queue-based Lyapunov function. The weight trades cost against queue growth. One source so far, a reinforcement-learning offloading paper, with author-reported simulation results only and several undefined terms.ConceptMachine learningGetting a computer to learn patterns from example data instead of following rules written by hand. Covers the three learning types, the four aspects of building a model, how models are assessed, and the practical risks: bias and drift in data, explainability and resource limits. Two sources: a conceptual review and an open textbook with runnable code. Neither reports experiments of its own.ConceptMachine unlearningRemoving a training point's influence from a trained model without full retraining. Exact deletion must match a retrained model; behavioural unlearning only stops forgotten outputs, and adversarial prompts can bring some back. Author-reported: k-means deletion 2.2x to 584x faster than retraining; certified linear removal 250x to 870x; LLM gradient ascent cuts harmful outputs to 1-6%.ConceptMagnetic actuationMoving a small body with an external magnetic field instead of onboard power. A rotating field makes a magnetized millirobot roll, flip, swim or steer, and a short pulse makes it jump. A haptic loop lets an operator feel a field-driven micro-robot. The trade-off: the body must respond to the field, and its motion follows the field.ConceptMahalanobis distanceA distance that uses the inverse covariance, so correlated or differently scaled features are not treated alike. Three uses: choosing similar past cases before fitting a local model (one clinical paper, where weighting helped Naive Bayes but not a Bayesian network), and two OOD scores, one on labelled class centroids (ViM baseline) and one on unlabelled self-supervised features (SSD).ConceptManipulation taxonomyHow robot manipulation is classified and counted. T-DOM tags each step of a deformable-object task by deformation, motion energy regime and contact; DexGrip compares gripper designs by degrees of freedom. Tags are hand-made with no agreement score, and DoF counts follow each author's convention. Author-reported.ConceptMasked image modelingMasked image modeling hides parts of an image and trains a network to recover them, with no labels. Variants differ in the target (pixels, discrete codes, an EMA teacher or prototypes) and in whether the encoder sees the hidden parts. Pixel targets are the cheapest; teacher and code targets cost more, and none wins on every protocol.ConceptMasked language modelingHide part of an input and train a model to recover it from the visible rest, with loss at the hidden positions. Covers encoder pretraining in several languages, table-and-text, tabular, fusion and label masking, cloze prompts, masked diffusion, iterative decoding, replaced-token detection, blank infilling, adversarial fill-ins and bias probes. The unit you hide is the main design choice.ConceptMaterial point methodA hybrid particle-grid method for simulating deformable materials: particles carry mass and deformation, a background grid solves the momentum update each sub-step. Used to make static 3D Gaussian objects move under forces, with an elastic material model. Only one card covers it so far; it fits Young's modulus from a generated video and costs about a minute per second of video on one V100.ConceptMathematical morphologyMathematical morphology uses grey-scale dilation, closing and masked fills, with a fixed kernel per step, to shape an image or depth map. One card: a classical chain that densifies sparse LIDAR depth with no training and about 90 Hz on a CPU. Its constants were tuned greedily on KITTI validation, so they may not transfer. All numbers are author-reported.ConceptMatrix completionRecovers a low-rank matrix from a random subset of its entries. Spectral completion rescales the observed entries by 1/p and takes a rank-r truncated SVD. It needs uniform sampling, incoherence and a known rank; the monograph gives subspace and Frobenius bounds, plus entrywise bounds for the same estimator.ConceptMechanistic interpretabilityFinds the internal parts behind a language model's behaviour with causal interventions. Path patching finds a 26-head circuit for indirect object identification in GPT-2 small. Other variants trace information flow, learn sparse features, read feed-forward value vectors, probe and steer a readout, or shift probe-chosen heads. Most need white-box access.ConceptMedical image synthesisGenerates a medical image from another modality, such as a CT volume predicted from an MRI of the same patient, or from a text prompt. Two cards: a patch-based 3D GAN trained on paired, voxel-aligned MR and CT, with metrics on 16 and 22 subjects against older baselines; and a dentistry review whose one diffusion example of synthetic X-rays has no measurement.ConceptMel-frequency cepstral coefficientsCompact per-frame audio features: a short-time spectrum, warped onto a mel or linear filterbank, log-compressed, decorrelated with a DCT, plus deltas. WaveFake uses MFCC and LFCC in a GMM deepfake detector: best MFCC aEER 0.214, best LFCC 0.062. Two emotion cards list MFCCs among their speech features, with no MFCC-only result. Author-reported.ConceptMembership inferenceAsks whether a record was in a model's training set. Covers shadow-model, per-record reference-model and distillation attacks, a black-box loss-threshold test, poisoning-assisted DP-SGD audits, and a sentence-level test on fine-tuned GPT-2. Numbers are author-reported. DP-SGD cuts measured GPT-2 leakage, but no card gives its utility cost.ConceptMembrane computingMembrane computing models a cell as compartments that split and merge. In metaheuristic search, the population is split into membranes that search separately and are then merged. One author-reported card so far: it beats its base optimizer on a benchmark, at about 1.7 times the runtime, and its ablations lose to the full method.ConceptMemory planning for inferenceGiving each intermediate tensor of a model an offset in one reused scratch buffer, so that tensors alive at the same time never overlap. Exact-cover packing of live-range rectangles can reach the minimum buffer size, while greedy first-fit can waste RAM. The only source so far is one microcontroller compiler, and its numbers are author-reported.ConceptMeta-learningTraining over many small tasks so a learner adapts quickly to a new one, or generalises across simulated shifts. On miniImageNet 5-way, a plain embedding with a per-episode linear probe reaches 62.02/79.64% (1/5-shot, author-reported). Twelve cards list the term, and they use it for different things. Main trade-off: where the gain comes from, the algorithm or the learned features.ConceptMetaheuristic optimizationPopulation-based search for hard, black-box objectives, simulator failures and feature subsets. Eight cards: a swarm method, a membrane optimizer, surrogate-based GA and NSGA-II, two simulator test generators, two feature-selection studies, and a learned router that uses a metaheuristic as its baseline. Trade-offs: tuning effort, no guarantees, author-reported evidence.ConceptMetric learningTrain an embedding so that matching inputs end up close and non-matching ones far apart, using pair, triplet, N-pair or proxy-based losses over a distance or cosine score. Pair and triplet losses need only similar/dissimilar labels but need hard-negative mining and converge slowly. Proxy-based losses avoid pair mining and need a class label per sample, plus a proxy count per class.ConceptMinibatch optimal transportEstimates an optimal transport loss on random minibatch pairs so it can train a model. Balanced minibatch OT forces every sample to move, which pairs unrelated samples; unbalanced variants let mass go unmatched, with unbiasedness and concentration proved. Gains in domain adaptation are author-reported; one paper so far.ConceptMinimax optimizationTwo-player problems with an outer minimization and an inner maximization. Two cards: a normalized gradient descent-ascent method with clipped inner ascent for heavy-tailed gradient noise (2509.14952), and an adversarial-training min-max whose inner attack is stopped by a stationarity score (2112.08304). Their results are author-reported and not comparable.ConceptMixed-effects modelsMixed-effects models add per-subject random effects to a regression, so repeated measurements from the same subject are modelled correctly. Tree and forest hybrids (MERF, RE-EM, GMERF, BiMM and relatives) replace the linear fixed part with a learned one. Trade-off: the random-effect assumptions must hold, and the fitting loops are heavier than a plain forest.ConceptMixed-precision quantizationGiving different tensors or layers different bitwidths or number formats instead of one width for the whole model. A compiler for microcontrollers picks 8- or 16-bit per tensor under a RAM limit, with a measured result of 54.81% against 55.02% float. A survey recommends FP16 for norms and end layers and INT8 for dense layers, but reports no ablation for that split.ConceptMixture of experts (MoE)A router sends each token to a few expert sub-networks, so a layer holds more parameters than it runs per token. Variants differ in expert placement, routing losses and overflow handling. A motion VAE blends six experts with a gate. Results are author-reported. A video model picks experts by noise level; one closed model report names MoE without numbers.ConceptMode connectivityWhether independently trained minima of a neural network's loss are joined by paths along which the loss stays low. The core idea is to search for a minimum energy path between two minima and report its highest loss, the saddle, minus the minima's loss, the barrier. The main trade-off is cost: each path needs a full AutoNEB run, and the result is only an upper bound on the true barrier.ConceptModel FLOPs utilization (MFU)Training-efficiency metric: measured tokens per second divided by the tokens per second peak matmul FLOPs would allow for the model's required forward and backward work. Recomputation is not credited. PaLM 540B reports 46.2% MFU vs 57.8% HFU. Needs an analytic FLOP count per token and a peak-FLOPs figure.ConceptModel editingChanging a trained model's answer to specific inputs without retraining it. Variants: fine-tuning with a locality penalty (baseline), learned gradient editors (MEND), memory-based editors (SERAC), and in-context editing. The trade-off is edit success against drawdown on unrelated inputs, and against weight changes or per-query cost. Weight editors often miss implied facts.ConceptModel inversionModel inversion recovers inputs from a trained network by optimizing noise until the model assigns a target class or embedding. A BatchNorm-statistics prior needs no training data (ArcFace); generative variants add a GAN or diffusion prior. A 2024 survey covers black-box, text and graph attacks plus defenses. Cost: many gradient steps or queries.ConceptModel mergingMerge models with one architecture into one, with no joint training run. Variants: matched averaging (FedMA), Fisher-weighted averaging, a closed-form merge of fine-tunes (Custom Diffusion, about 2 s), SLERP/TIES/DARE on LLM weights (BioMistral), or PAINT, which blends a CLIP model with its one-task fine-tune. A shared-subspace projection compresses many fine-tunes instead.ConceptModel predictive control (MPC)Solve a finite-horizon optimal control problem from the current state, apply only the first input and re-solve. In autonomous racing it is the most common control approach. Variants add nonlinear, sampling, learned or logic-based models, learned or imitation-trained policies and LLM-set parameters. Racing sources give no head-to-head comparison.ConceptModel routingSends each query, or each step of a task, to one of several models: a router picks it, or a trained planner writes the plan. A survey covers per-query routing as an efficient-inference family; a 2025 paper trains a 7B planner over a worker pool; a GPT-5 system card describes a deployed router with no features given. Results are author-reported.ConceptModel selection without target labelsChoosing which trained model to keep when the usual hold-out check needs labels you do not have, such as labels from the new domain under unsupervised domain adaptation. Options are source risk, importance-weighted validation and feature-space Deep Embedded Validation, which the authors report comes close to target-label selection. One paper so far.ConceptModel-based reinforcement learningChoosing actions with a model of the environment: plan over a model you estimate or learn, or train a policy on imagined rollouts or evolved controllers. Exact planning is cheap only for small state spaces. Learned world models cover several domains; one reported setup needs up to 16 GPU days per benchmark.ConceptModeration analysisTesting whether one variable's effect on an outcome changes with a third, the moderator. One survey method is OLS regression with interaction terms of mean-centred scores, entered in blocks. One e-commerce chatbot survey (n = 299) shows it. Trade-off: conditional effects from cross-sectional data, but no proof the moderator causes the change.ConceptMolecular representation learningTurning a molecule into a vector a model can use. Variants: fixed or learned fingerprints, RDKit descriptors, CNNs and RNNs over SMILES, Transformers over SMILES or substructure tokens, 2D message passing, and 3D equivariant networks. Generative 3D models such as GeoDiff output atom coordinates instead. The sources test different tasks, so their numbers are not comparable.ConceptMonocular depth estimationPredicting a value for every pixel of one RGB image: scene depth, or height in aerial imagery. Variants differ in training data, output head, loss and scoring. Omnidata, Marigold and HR-Depth report aligned depth; AdaBins, BinsFormer, NeW CRFs and Metric3D v2 report absolute errors on NYU and KITTI. The best-known Omnidata data and weights are non-commercial only.ConceptMonotonic alignment searchFinds a monotonic, surjective alignment between text tokens and spectrogram frames by maximising the frames' likelihood under each token's predicted mean, so no external aligner is needed. Known from one paper card, Grad-TTS; its search procedure is not described there.ConceptMonte Carlo dropoutKeep dropout on at prediction time and average several stochastic forward passes; their spread stands in for model uncertainty without retraining. Active learning uses it for BALD-style scores. Trade-offs: several passes per sample, overconfidence in a regression tutorial, too-large variance inside the data in a toy regression, and no guaranteed win over softmax confidence.ConceptMonte Carlo tree searchMonte Carlo tree search builds a look-ahead tree from simulated futures and picks moves with the UCT rule, which balances mean value against rarely tried options. Variants swap random playouts for learned evaluators (AlphaGo, AlphaZero), search over particle beliefs, search over LLM agent actions, or search imagined robot futures (STORM). Trade-off: many simulations or model calls per decision.ConceptMotion predictionPredicting where people or interacting agents move next from past motion, or a listener's face from a speaker's words. Six variants: a recurrent plan-and-body model, a transformer that outputs a block of future frames, an equivariant interaction model, a discrete-token language model, a joint latent model of traffic, and forecasts fed into a social planner. Reported numbers use different metrics.ConceptMotion primitivesShort pieces that carry a system from one state to the next. Constant-input pieces with a closed-form cost form graph edges for quadrotor A* planning; learned pieces are conditioned on a partial observation for humanoid teleoperation; parameterised skills are checked inside an LLM planner. Results are author-reported.ConceptMotion retargetingMapping motion from one body onto a robot whose joints differ. Variants: per-step keypoint or fingertip-vector optimization for dexterous hands, frame-wise or clip-level IK with feasibility filtering or smoothing for humanoids, task-space QP with a fixed tracker-to-joint assignment for VR teleoperation, and a learned human-to-cobot mapping refined by IK.ConceptMulti-agent reinforcement learningSeveral learning agents act in one environment. Independent PPO with shared weights matches or beats joint-value methods on several StarCraft maps. TAR² splits a sparse team return into per-agent, per-step rewards. A per-base-station DDPG allocates edge resources. Results are author-reported; this page has four card-backed sources.ConceptMulti-agent systemsLanguage-model agents with roles and tools split a goal and coordinate by messages, shared memory, debate, a planner or tree search. Twenty-eight card-backed papers cover software, robots, medicine, biomedical and materials work, debate, peer review, causal discovery, simulation and surveys. Gains are mostly author-reported; some are within noise.ConceptMulti-armed banditSequential choice among K arms with unknown mean payoffs. Covers the trade-off between cumulative reward and accurate estimates of every arm, a weighted objective that balances the two (ForcingBalance), and a bandit task used to test whether GPT-3 explores like people. Author-reported results only.ConceptMulti-criteria decision makingLearn a user's ranking of mined patterns as a weighted sum of interestingness measures, with weights from the Analytic Hierarchy Process (AHP) over pairwise agreement gaps. Passive mode learns from rankings; active mode asks pairwise questions. Trade-off: much faster than an SVM ranker, but less accurate in passive mode. Author-reported.ConceptMulti-hop reasoningMulti-hop reasoning answers a question by chaining two or more facts. Benchmarks, prompting, interleaved retrieval, document-link pretraining and looped transformers approach it differently. Score the answer and the evidence chain together: one baseline reaches 5.41 joint F1 against 43.93 answer F1 (author-reported).ConceptMulti-label classificationPredict a set of labels for one input, where labels are not mutually exclusive. The default is one sigmoid per label with binary cross-entropy. Variants handle uncertain targets, weight noisy candidate labels, use known labels at test time, change the loss, or give one softmax per hierarchy depth. Trade-off: more compute, and for one token per label, cost that grows with the label count.ConceptMulti-layer perceptron (MLP)A feed-forward network of stacked layers, each a weighted sum followed by a nonlinearity. Twenty-one cards use it or test it against an alternative, from anomaly detection and choice models to click prediction, forecasting, pixel classification, terrain sensing, exoskeleton torque and microcontrollers. Most trained settings are unstated or library defaults.ConceptMulti-person pose estimationFind each person in an image and their body keypoints, each keypoint assigned to its person. Families: bottom-up grouping, top-down detect-then-estimate, and single-stage keypoints on a detector head. Cards cover one bottom-up method built for CPUs and one single-stage method. Latency, compute, precision and the reported metric decide the choice.ConceptMulti-query attention (MQA)Every query head shares one key head and one value head, so the KV cache shrinks by the head count and decoding reads far less memory. PaLM uses it and reports neutral quality (no MQA ablation). On T5-XXL, uptrained MQA averages 0.6 points below MHA at 0.24 s vs 1.51 s per sample (author-reported).ConceptMulti-robot navigationSeveral robots share one space and must reach goals without collisions or standoffs. Two variants: a reinforcement-learning policy for drones that explore an unknown room and carry a payload (simulation only, under-specified), and a neural controller that imitates an MPC and adds a control-barrier-function QP so two robots pass doorways without deadlock (simulation only, code released).ConceptMulti-scale feature fusionMerge feature maps from several levels so one prediction sees fine detail and high-level meaning. Choose the levels, the size and width match, the merge (concatenation, sum, learned gates) and how the levels are built. FSSD fuses in one step (+1.6 mAP over SSD300 in its table, but slower). Dynamic Head gates levels per input. Segmentation and other tasks fuse skip paths.ConceptMulti-task learningTrains one model on several objectives at once. Designs: a shared encoder with per-task heads and a summed or weighted loss; one loss where a prompt, mask or mixture sets the task; or one update that merges task gradients. Twenty-six cards show these. Joint-versus-separate results disagree, and loss weights vary (1.0, 0.3/0.4/0.3, equal, adapted, or not stated).ConceptMulti-view clusteringClustering samples with several feature sets (views) into one partition. FastMICE ensembles anchor-based graphs built from random view groups. CPM-Nets learns one code per sample from its observed views, so views may be missing. Author-reported: best average NMI rank on 22 datasets, and ACC and NMI gains on six partial-view sets. Trade-offs: k must be given; released code differs from the papers.ConceptMulti-view reconstructionRecovering a 3D surface from posed images, by fitting a representation through rendering or by predicting a volume directly. NeuS fits a neural SDF (DTU Chamfer 0.77). 2D Gaussian Splatting fits oriented disks (0.80 in about 11 min). NeuralRecon predicts TSDF fragments from video (30 ms per key frame). One-2-3-45 and One-2-3-45++ make a mesh from one photo in 45 to 60 s.ConceptMulti-view representationRepresent a 3D scene by several fixed rotated views, give each object a shared feature plus a per-view position code, then average the fused outputs so the result does not depend on the starting view. One card so far, for 3D visual grounding. All numbers are author-reported on Nr3D and Sr3D.ConceptMulti-view stereoEstimating depth from calibrated photos by building a cost volume over depth hypotheses. Four variants: UniMVSNet (slice-plus-offset read-out, author-reported first on DTU); Pseudo-LiDAR++ (depth-trained stereo with sparse LiDAR correction, for 3D detection); Point-NeRF (MVS-seeded neural point cloud); MVSplat (one-pass Gaussians from a plane-sweep volume).ConceptMultibody dynamics of closed-loop mechanismsMotion equations for parallel robots whose limbs contain closed kinematic loops. Each loop is solved locally, so the limb acts as a serial chain, and the robot gets one task-space model. Two author-reported variants: a screw/Lie-group model with a linearised loop map, and a numerical loop solver with Newton correction. Each is shown on one example, with no error figures.ConceptMultilingual evaluationMeasuring how models behave across languages and cultures. Belebele gives parallel items in 122 variants. Other designs score many-language tasks by label likelihood, probe Arab versus Western entity preference, simulate survey respondents, test one country's answers, run code in five programming languages, or score word vectors on new pair lists. Each asks something different. Author-reported.ConceptMultimodal fusionCombining inputs of different kinds, such as text, audio, video, touch, depth, radar, images or lab values, into one output. Twenty-one cards describe designs, from shared codes and joint transformers to depth gates and evidential rules. A survey, a benchmark, a dataset and a position paper complete the page. The numbers come from different tasks, so they do not rank the designs.ConceptMultimodal language modelA language model that reads images, video, audio or other signals next to text and answers in text. Designs differ in how the non-text input enters the model, which parts train, and whether the output can also be masks or boxes: a frozen LLM behind a small bridge, a natively multimodal decoder, a fully fine-tuned generalist, or one token vocabulary for everything.ConceptMultiple instance learningMultiple instance learning trains a classifier on bags of instances that carry one label per bag, such as a whole-slide image split into patches. Pooling methods combine instance scores with a fixed rule. Correlated variants let instances attend to each other with a Transformer, which needs an attention approximation for bags of thousands of patches.ConceptMultiplicative weights updateThe multiplicative weights rule: each action's weight is multiplied by its expected utility and the weights are renormalised. Covers the parameter-free and mixed-utility variants, when each applies, and what regret guarantees follow. Based on one paper so far, which links the rule to sexual population dynamics.ConceptMultivariate normal integrationComputes the probability that a normal random vector lies in a region. Use a generalized chi-square route when the boundary is quadratic, and a ray-trace integral over directions for any other shape. Monte Carlo sampling is the simple baseline, but it loses accuracy once the probability is tiny.ConceptMutual information maximizationAdd a loss term that rewards a learned representation for sharing information with another variable. Mutual information is hard to compute exactly, so each variant uses a bound or estimator. One card so far: a Barber-Agakov bound on input pairs plus a contrastive term on the fused vector, for multimodal sentiment. Author-reported gains are small; the bound needs extra predictors and a memory.ConceptNamed entity recognitionFinding and typing entity mentions such as people or organisations. Covers trained taggers (CRF, encoders and domain encoders), token and model-based augmentation, LLM prompting for spans and PHI redaction, LLM pre-labelling for experts, and tagger-based scrubbing of training text. Trade-off: prompting suits small labelled sets but trails supervised taggers on benchmarks and costs API calls.ConceptNatural language inferenceDeciding whether a premise supports, contradicts or leaves open a hypothesis, and how that task is built and checked. Two cards so far: a hypothesis-only baseline that exposes label cues in SNLI and MultiNLI, and the fact-checking survey's view of entailment as a verification step. Both are reports, not new systems.ConceptNearest-neighbor searchFinding the k stored points nearest each query under a distance. Nine cards: exact GPU brute force and fused reductions, exact imitation-learning lookup, approximate transformer memory, Faiss compressed indexes, RETRO retrieval, PaCMAP's neighbour graph, and an archive density test. Author-reported numbers use different hardware and do not compare directly.ConceptNested case-control samplingReplaces the full risk-set sum in a survival likelihood with one control drawn per event, so the loss can be trained on mini-batches. The paper reports that one control stays close to the exact likelihood on simulations, and that more controls lower training error at higher compute cost.ConceptNetwork pruningRemoves the least useful parts of a model by a scoring rule, or rewires a sparse model during training, so it gets smaller or cheaper to run. Variants differ in unit (weights, kernels, filters, LLM groups, search edges, 3D primitives, neurons) and score (L1 magnitude, gradient, posterior variance, learned mask). Finer units keep more accuracy; coarser ones give regular structure hardware can use.ConceptNetwork resilienceMeasures how well a network withstands a failure and how far it recovers, from a performance curve over time. Shows three numbers (lowest point, value at a cutoff, time average), a per-node risk SIRD failure model, and protection or recovery actions triggered by a forecast. One card, simulation only, with author-reported gains on a single traffic network.ConceptNeural ODEWrites a deep stack of layers as a continuous process described by an ordinary differential equation, solved numerically. Only one card so far: a graph ODE with a restart term that keeps node states from blurring with depth, used for traffic forecasting. Choose it when the data sits on a fixed graph and deep graph stacks over-smooth; check the solver steps, which the paper leaves open.ConceptNeural architecture searchAutomatic design of network structure: a search space, a strategy for proposing candidates and a cheap way to score them. Evolution and RL cost hundreds to thousands of GPU-days; weight-sharing and gradient methods report under a few GPU-days on CIFAR-10. Trade-offs: search cost, a gap between searched and deployed depth, and shared weights that can distort rankings.ConceptNeural combinatorial optimizationLearning a policy that builds routes or other discrete solutions one decision at a time, trained by reinforcement learning. One card so far: an attention-based pickup-and-delivery route builder. Author-reported: faster early answers than OR-Tools, close final quality only up to 200 stops, and hours to days of training per model.ConceptNeural implicit surfacesNeural implicit surfaces store shapes as learned scalar fields and read the surface from a level set. NeuS fits a signed distance, and UNISURF fits occupancy. PhySG adds material and light. CLIP-Forge decodes occupancy from text, iSDF fits online, One-2-3-45 predicts one from one photo, and NeuralRecon, a discrete exception, outputs a voxel TSDF per fragment.ConceptNeural machine translationNeural machine translation (NMT) maps a source sentence to a target sentence with one trained network. Covers BPE subwords, a speaker-gender tag, speech-to-text translation, BARTScore, GPT translation and judging, MQM annotation, NLLB-200 and an LLM translate-then-classify pipeline. Results are author-reported; gender-tag BLEU gains are under 1.5 points.ConceptNeural operatorLearns a map from an input function, such as PDE coefficients or a blade shape, to an output field, so one model is not tied to one grid. Cards: a multiwavelet operator that beats FNO on PDE tables, a branch-trunk operator with lower field error than FNO on one compressor dataset, and DeepONet uncertainty bars. Gains depend on the baseline; finer grids lose accuracy.ConceptNeural radiance field (NeRF) and variantsA neural radiance field stores a scene in a learned function of position and view direction, rendered by volume rendering along camera rays. Variants trade render speed, training cost, surface quality and few-view robustness: KiloNeRF renders fast with tiny MLPs, Plenoxels trains fast with no network, NeuS recovers surfaces slowly, and DS-NeRF and FreeNeRF help with few views.ConceptNeural scaling lawsHow to fit a power law of error against a training resource, forecast larger runs, and see where the forecast breaks. Covers CLIP runs, published-compute meta-analysis, data filtering, a small model on curated data, a learning-rate schedule, and size, task or data sweeps with no fit, including a robot diffusion policy. Forecasts beyond the measured range are extrapolations.ConceptNeural style transferRestyles images and meshes. ArtFlow, a reversible flow against content leak (SSIM 0.45 against 0.29 for AdaIN); CLIPstyler, from a text prompt alone (3.78 against 3.32); STROTSS, per-image optimal-transport optimisation (95 s per 512 px image); Splice, a per-pair network on frozen DINO features (79.0% against STROTSS); Text2Mesh recolours a mesh (3.90 against 2.83).ConceptNeural-symbolic integrationNeural-symbolic integration joins learned representations with explicit symbols and logic, aiming for systems that learn, reason and explain. Fourteen papers map the options, from surveys and program-writing language models to a solver-free loop, a KAN, a mental-state review, a robot method and a logic tensor network. Method results are author-reported. Pick a coupling pattern by constraint.ConceptNeuroevolutionEvolving networks with gradient-free search. Topologies scored with one shared weight, no weight training, beat fixed baselines on three control tasks. A linear controller evolved with CMA-ES over a learned world model reaches 906 on car racing. Co-evolved bodies with PPO learners learn new tasks faster. Author-reported. Trade-offs: large search budgets, fixed-weight scoring, feed-forward only.ConceptNo-regret learningA learner's regret is the gap between its cumulative payoff and the best fixed action in hindsight; no-regret learning means that gap shrinks over time. Covers the rescaled-game bound with step size equal to the selection strength for each gene, the recombination-before-selection variant, and what regret does not guarantee (fitness, convergence). Based on one paper so far.ConceptNode embeddingMaps every node of a graph to a fixed-length vector so that ordinary classifiers and regressors can predict node properties. Four families in the sources, from random-walk skip-gram to a GNN inside a language model. Evidence is four author-reported papers on different tasks, plus a survey with no numbers, so the families cannot be ranked against each other.ConceptNoise residual featuresHigh-pass residuals (fixed SRM filters) that remove colour and keep fine noise, fed as a second stream to spot manipulated faces. Only one paper card so far: a two-stream detector whose ablation shows the noise stream alone helps on unseen forgery methods but not on the training one, and that fusion and multi-scale features add to it. All numbers are author-reported.ConceptNoise scheduleThe noise schedule sets how much noise each diffusion step adds, usually as gamma(t) = -log SNR(t). Variants: a learned monotone schedule on the variational bound, a learned Gumbel schedule for text, truncated and adaptive schedules for flow LMs, zero-terminal-SNR rescaling, and fixed DDPM or square-cosine tables. Results are author-reported.ConceptNoisy channel modelScores each label by how well it explains the input, P(x | label), instead of P(label | x). With frozen GPT-2 Large on 11 text classification datasets and 16 examples, the author-reported channel prompt tuning averages 61.7% against 57.7% for direct head tuning.ConceptNonprehensile manipulationMoving an object without a closed grip, here by tilting it against two supports with a curved palm and then tucking the palm under it. Only one card so far, so the page covers one method. Its contact order decides whether the tilt is feasible, and planning uses a force-closure test over (theta, delta). Author-reported, with 5 trials per setting.ConceptNormalizing flowsNormalizing flows are invertible maps with a tractable Jacobian, so density and inverse are both exact. Five cards: a continuous flow that edits StyleGAN latents, a coupling flow for style transfer, a conditional RealNVP prior for text-to-shape, a conditional flow for anomaly maps, and a survey of the families. Trade-off: exact invertibility against compute per step.ConceptNovel category discoveryLabelling unlabelled data that may mix known and new classes when the number of classes is unknown. One paper so far: Generalized Category Discovery clusters contrastively fine-tuned ViT features with semi-supervised k-means, pins labelled images to their classes, and picks k by labelled accuracy. Trade-off: choosing k needs labels, and fine-grained or long-tailed data scores much lower.ConceptNovel view synthesisNovel view synthesis renders unseen views by fitting a field or Gaussians. KiloNeRF, K-Planes and HexPlane use grids or tiny MLPs. DONeRF places 2 to 16 samples per ray. FreeNeRF, DietNeRF and FSGS suit few views. BARF fits poses. pixelNeRF, FLARE and Splatter Image skip fitting. SV3D orbits one image. A Gaussian tree covers cities. Forge4D, PSHuman and DriveSplat cover people and driving.ConceptNuclear-norm regularizationAdds the sum of a matrix's singular values, the nuclear norm, as a penalty on a learned matrix, which pushes it toward low rank. Covered: the exact norm and the top-K estimate used for a per-layer graph mask. The trade-off: the exact norm needs a full SVD, the top-K estimate adds a bound and a K to choose, and the only source is one graph paper with author-reported gains.ConceptNucleus (top-p) samplingNucleus (top-p) sampling picks each next token from the smallest set of most likely tokens whose probability reaches p. Its factual variant shrinks p inside each sentence to cut invented facts, at a cost in diversity and repetition. Protein models and a consensus pipeline also use it. Three paper cards; all numbers are author-reported.ConceptNudged elastic bandFinds a minimum energy path between two minima of a network's training loss by moving a chain of parameter vectors (pivots) so that only the loss force across the path and the spring force along it act. AutoNEB adds pivots where the path is poorly sampled. Reported saddles stay near the minima's loss on CIFAR models, but they are upper bounds from local search.ConceptObject detectionFind the objects in an image and give each a class, a box and a score. Variants: two-stage region proposals, single-stage dense detectors, DETR-style set prediction, label assignment for tiny objects, and 3D LiDAR or BEV detection. Latency, hardware, data and training budget decide the choice. The papers use different datasets and metrics, so their numbers cannot be ranked.ConceptObject queries (learned query slots in transformer decoders)Query slots that a transformer decoder turns into outputs in parallel: boxes in DETR, 3D boxes in DETR3D, points in CLTR, class masks in Segmenter, 3D instance masks in Mask3D, depth bins in BinsFormer. Variants add a reference point or a sampled position per slot, carry slots across frames or match whole sequences. Slots meet targets by matching, a fixed class map or a pre-assigned positive.ConceptObject-centric learningLearn one representation per object from unlabelled inputs. Variants: EM-style iterative pixel grouping (N-EM) and its recurrent version for sequences (RNN-EM); a survey of binding mechanisms with no experiments. MOVi video benchmarks (from Kubric) measure object discovery. Author-reported numbers are on synthetic data, so they do not show real-image behaviour.ConceptOffline reinforcement learningLearning a control policy from a fixed log of past experience, without new environment interaction. Variants keep the policy near logged actions in different ways: TD3+BC adds a behaviour-cloning term, DeFlow learns a residual on a frozen flow prior, QPHIL plans over learned landmarks. Gains are author-reported, on different benchmarks.ConceptOn-device inferenceRunning a trained model on the device that needs the answer, from microcontrollers to phones and Jetson-class boards, instead of on a server. The core idea is to compress or redesign the model to fit memory, compute and power limits, then port it to a device runtime. The trade-off: small models lose capability, and reported speed and energy figures rarely carry over to other hardware.ConceptOn-device trainingTraining or adapting a neural network on the target device itself, e.g. a Cortex-M microcontroller with little RAM. Fully quantized training runs forward, backward and update in 8-bit, and per-sample sparse error updates cut backward work. One paper so far; its numbers are author-reported and the best accuracy needs the last layer kept in float.ConceptOn-policy distillationA student writes its own rollouts and a teacher scores each token, a dense signal instead of one outcome reward. Variants differ in how they use the teacher's probabilities, and a reverse-KL baseline cuts the variance of the one-token signal. Trade-offs: teacher probabilities on every rollout, and a teacher choice that depends on student size and task. Based on three papers.ConceptOne-class classificationFlags inputs that do not belong to the single class seen in training, using inlier data only. Covers two-stage detectors that fit a kernel density or one-class SVM on self-supervised features, and separated novelty rejectors that filter inputs before a predictor. Image numbers are author-reported, and the paper's tuning used outlier images.ConceptOpen-set recognitionClassify known classes and reject inputs from classes never seen in training. One card: a stronger closed-set classifier scored by its maximum logit matches or beats prior open-set methods on standard benchmarks, and is within a few points of ARPL+ on a fine-grained semantic-shift benchmark. All numbers are author-reported.ConceptOpen-vocabulary segmentationName each pixel or 3D point with a class given at test time. Covered: OpenSeg (72.2 PASCAL-20 mIoU), GroupViT (52.3 VOC 2012), OVSeg (94.5 PASCAL-20), ZSSeg (37.8 COCO-Stuff zero-shot hIoU), SAN (27.5 ADE-150), Feature 3DGS (0.782 Replica), ConceptFusion (0.446 3D mIoU, UnCoCo) and SparseGrasp (55/70 grasps, 3 views). Numbers are author-reported, with different backbones and data.ConceptOpinion dynamicsOpinion dynamics predicts how a group of agents that hold stances changes them through exchange, and which collective state results. The card covers 32 persona-conditioned LLM agents on a signed graph, fitted by an Ising-style logistic rule with separate concordant, discordant and unsigned couplings. The rule is author-reported at 75-86% one-step balanced accuracy on held-out questions.ConceptOptical character recognitionReading the text that appears in an image. Two strands: FAN, a scene-text recognizer whose focusing network pulls attention back onto each character, and OCRBench, a 1000-question zero-shot test of text reading in large multimodal models, scored by whether the ground-truth string appears in the reply. Every reported score is the authors' own.ConceptOptical flowDense 2D motion between two frames: an estimator's output, an alignment signal in a video model, or a training target. Estimators are global matching (GMFlow, Unimatch) and a transformer over the cost volume (FlowFormer). Trade-offs: memory grows with the square of the pixel count, and alignment follows the flow. Author-reported.ConceptOptimal transportCheapest way to move mass between distributions under a cost such as squared distance. Eight uses: an input-convex-network map, unbalanced minibatch OT as a loss, Sinkhorn matching in a feature matcher, Sinkhorn prototype clustering, a relaxed EMD style loss, a 1-D distance for motion scoring, a key-to-finger reward, and a temporal OT reward. Author-reported numbers only.ConceptOriented object detectionDetect objects as rotated boxes (x, y, w, h, angle), mainly in aerial images where objects have arbitrary heading. Two variants: a rotation-equivariant backbone with rotated RoI pooling (ReDet), and an end-to-end transformer with rotation-aware attention (OrientedFormer). Choose by whether you can change the backbone, your compute budget and whether you want to avoid NMS.ConceptOrigami mechanismsA flat sheet folded along set creases gives one body several jobs: a folded cylinder can roll, flip, swim, pump liquid by folding and grip by spinning. Its geometry sets the locomotion modes and its cavity sets the storage. The trade-off: the folds set the function, so changing the job means a new fold pattern or a second part. Only one paper covers it so far.ConceptOut-of-distribution detectionFlagging inputs a model was not built for. Post-hoc scores (ViM, max logit) leave training alone; SSD learns features without labels; LogitNorm changes the loss; MOS groups classes for large label spaces. Author-reported: ViM AUROC 90.91 (BiT); LogitNorm CIFAR-10 FPR95 49.52 to 15.65; MOS FPR95 39.97 vs 54.30 (KL matching).ConceptOutlier rejectionRemoves wrong matches or measurements before a geometric fit, so the final estimate uses only inliers. Learned spatial-consistency pruning (PointDSC) reports 93.28% registration recall on 3DMatch at 0.09 s per pair, against 5.50 s for RANSAC-100k (author-reported). The main trade-off is the inlier threshold, which must be set per dataset.ConceptParallel scan (prefix scan)A parallel scan evaluates a recurrence over a whole sequence at once by combining steps with an associative operator, so the result matches the step-by-step loop. Two variants: S5, a diagonal state space layer run in the time domain at S4's cost, and a scan over LQR value functions in GPU model predictive control for legged robots. Both are author-reported.ConceptParameter-efficient fine-tuningAdapt a pretrained model by changing a small part of it and keeping the rest frozen, so each task stores megabytes, not a full copy. Variants train scaling vectors, residual adapters, sparse masks, tensor factors, unlearning layers, random subspaces, paired or per-layer prompts, or control adapters; Tip-Adapter sets its few-shot CLIP cache without gradient training.ConceptPart-based featuresSplitting per-patch features into groups, pooling each group into its own vector and comparing those vectors. Two papers so far: FSRA orders patches by mean activation and cuts the order into 3 regions (drone-to-satellite); Amir et al. cluster frozen DINO ViT keys into parts with no training. Author-reported results on different benchmarks.ConceptPartial information decompositionSplits what several inputs say about a target into unique, redundant and synergistic parts. Sensory PID applies it to audio and video answers of omni-modal models, estimated from answer distributions with one stream masked, and can weight training samples.ConceptPartial label learningTraining when each example carries candidate labels, some of them wrong. Options: fixed weighting rules (hard, soft, identity), a calibrated per-candidate weight from a pre-trained reliability estimator, and an added prototype loss for high noise. Trade-off: no extra stage versus a calibrator run that is reused across datasets. Only multi-label image work has a card so far.ConceptPartial multi-view learningLearning from samples where some views (sensors or feature sets) are missing, without grouping samples by missing pattern or imputing first. CPM-Nets learns one shared code per sample from its observed views, with a class-structure loss. A GAN variant imputes missing views for unsupervised use. Results are author-reported; only one paper is covered so far.ConceptPartially observable Markov decision process (POMDP)Sequential decision making when the state is hidden and actions and observations are noisy. The agent plans over beliefs, probability distributions over states. Exact solutions are intractable, so practical solvers plan over a sampled set of beliefs (offline point-based methods such as SARSOP, or online tree search such as POMCP and DESPOT).ConceptParticle swarm optimizationSwarm search in which agents move by a velocity term. The wiki has three cards: a PSO-based variant that replaces velocity with a fitness-derived step, a survey that runs PSO as a feature-selection wrapper, and a survival paper that uses PSO to tune hyperparameters. No PSO-origin card exists, so baseline coefficients are not stated and all results are author-reported.ConceptPassage rerankingRe-scores a short list of retrieved passages with a second model before one is used as context. One card so far: In-Context RALM, where a frozen LM scores the candidates with no training, or a trained RoBERTa reranker picks one. Zero-shot reranking is reported on five corpora, the trained reranker on WikiText-103 only, and the trained variant needs 300,000 examples and a second model.ConceptPermutation language modelingPretrain autoregressively over a randomly sampled factorization order instead of left to right, so each token learns from context on both sides without [MASK] tokens or independence assumptions. The order lives only in the attention masks; a two-stream attention makes predictions aware of the target position. Trade-off: extra attention work and harder optimization versus BERT-style masking.ConceptPersonalized federated learningFederated learning where each client gets its own model. Three designs covered: a server-side hypernetwork maps a learned client embedding to that client's weights (pFedHN); a cloud model is fine-tuned per user with a CORAL term (FedHealth); a personal model is pulled toward the global one (Ditto). Results are author-reported.ConceptPerspective-n-point (camera pose from 2-D/3-D correspondences)Perspective-n-point (PnP) recovers a calibrated camera's rotation and translation relative to a known 3-D object from n matched points or lines. Two cards: a dual-quaternion trust-region refiner with point and line variants, and PnP inside RANSAC on learned 2-D/3-D matches (OnePose). Author-reported: at noise 1.00 px, translation error is 0.0002 cm against 11.48 cm for the linear method.ConceptPhysically based renderingPhysically based rendering computes colour from a material model and light, or traces rays through a scene. Four cards cover night Gaussian-splat shading, SDF inverse rendering for relighting, raindrop ray tracing for paired data and Kubric's path-traced scene generator. Code is released for PhySG and Kubric only.ConceptPhysics-informed neural networkA neural network trained so that its output satisfies a differential equation: the loss adds the equation residual and the boundary conditions. Some variants build a thermodynamic law into the output. Variants differ in point weighting, derivative estimation, boundary enforcement, penalty form and error bars. An SNR diagnostic tracks stalled training. Numbers are author-reported.ConceptPiecewise constant curvature (PCC)Piecewise constant curvature models a soft robot segment as a circular arc with one curvature, and maps tendon lengths to its length, bending plane and curvature. Kinematic PCC is explicit and fast; mechanics-based PCC adds static balances; multisection designs add tendon decoupling. Trade-off: cheap and explicit, but valid only when each section bends into an arc, and it does not capture torsion.ConceptPoint-to-grid interpolation (trilinear gridding)Scatters each point of a cloud onto the eight corners of its grid cell with trilinear weights, so a regular 3D grid can feed ordinary 3D convolutions. Dividing each vertex sum by its neighbour count, as the paper writes it, keeps values from growing with local point density; the released code sums without the count. Only one paper card covers this concept.ConceptPolicy gradientTraining a policy by raising the probability of sampled actions in proportion to a return or advantage, so the reward need not be differentiable. Variants differ in the baseline: a learned critic with GAE, a group of rollouts per prompt, a running mean, a greedy rollout, or none. Costs: extra rollouts, a critic, or a reward model queried at every step.ConceptPose-graph optimizationPose-graph optimization adjusts robot poses so that odometry, loop-closure and sensor constraints agree. Kimera (2101.06894) gives a pose-only back-end, a joint pose and mesh graph and per-person graphs. A continuum-robot sliding-window factor graph (2510.26623) is a related form without loop closures. Timings are author-reported on different hardware and problems.ConceptPosit number formatPosit is a bit-level number encoding that can replace float or fixed-point storage for model tensors. One wikillm paper uses it: MinUn stores each tensor as 8-bit or 16-bit posits on Arm microcontrollers, fixes the exponent size per width by accuracy, and reports lower RAM than float16 and bfloat16 at close accuracy. Only that one source is covered so far.ConceptPositional encodingHow attention, graph and generative models learn where each input sits, since attention alone ignores order. Variants: sinusoids added to the input, a flow condition on location, date and band codes, PPEG convolutions, rotary, distance and relative-bucket biases, box and camera-ray geometry, NeRF bands, an acyclic time code, and a learned node stream.ConceptPost-training pruningRemoving weights from a trained model in one pass, with no retraining. The one covered method, SparseGPT, picks a mask per layer from a small calibration set and corrects the surviving weights with a Hessian-based update. It reaches 50% unstructured sparsity on OPT-175B with little perplexity loss (author-reported). Magnitude pruning at the same sparsity collapses.ConceptPost-training quantizationConverting a trained network to integer weights, and optionally activations, without retraining. Per-channel weight ranges keep ImageNet CNNs near float accuracy; learned rounding (AdaRound) helps 4-bit weights. For LLMs, outlier features call for a 16-bit path (LLM.int8()), smoothing into the weights (SmoothQuant), or weight-only 3- to 4-bit rounding (GPTQ).ConceptPreference optimizationFine-tuning a model on preference judgments without a separate reward model or an RL sampling loop. DPO uses one logistic loss on pairs against a frozen reference and matches or beats PPO-based RLHF on sentiment and TL;DR summarisation up to 6B. KTO needs only a good or bad label per output. A diffusion variant applies the pairwise idea to video.ConceptProcessing-in-memoryPlacing computation inside or beside the memory that holds the data, so operands do not cross the off-chip DRAM bus first. Two placements are described: logic under stacked DRAM, and bitwise operations inside the DRAM array. Only one reuse-licensed card covers it, a survey with no PIM measurements, so treat the designs as a menu and measure on your own hardware.ConceptProduct of exponentialsProduct of exponentials (PoE) writes a body's pose on a serial or tree chain as matrix exponentials of joint screws times a reference pose. Covers the path product with fixed screws, a body-frame form with reference transforms, and the loop-aware use for parallel robots, where each closed loop is cut to a tree. Both sources are author-reported and use a single robot example each.ConceptProgram repair (LLM self-debugging)A frozen language model fixes its own program by running it, explaining it or both, then rewriting it, for up to 10 turns. No training. Gains are largest when tests exist (TransCoder with Codex 80.4 to 92.5). Without tests, explaining helps. For text-to-SQL, one zero-shot self-check adds 0.9 to 2.6 points on Spider dev. Needs code you can execute or check.ConceptPrompt engineeringSteering a frozen language model through its input: instructions, demonstrations, formats, reasoning cues, personas and layouts, and with weight access, learned soft prompts. Format changes alone moved one model's accuracy by up to 76 points; a zero-shot prompt tied 10-shot on one translation model. Evidence is mostly from 2021-2025 models, often one prompt per task.ConceptPrompt injectionText planted in an LLM application's input can override the developer's instructions. A survey names attack families and defences without numbers. Two system cards report author-run results on instruction-priority and injection tests, on their own test sets. A review of agent systems names the threat, with no experiments.ConceptPrompt optimization (LLM as black-box optimizer)Searching for a better prompt by scoring candidates on labelled data. OPRO uses a language model as the proposer and feeds the scored history back; its best GSM8K instruction scores 80.2 vs 71.8 for "Let's think step by step." (test, author-reported). A 2023 survey maps other methods (textual gradients, rewriters, tree search, RL) but runs no experiments of its own.ConceptPromptable segmentationSegment the structure a user points at, given clicks or boxes. Carded medical models: zero-shot SAM, where one box per part averages 0.6542 IoU, and SAM-Med3D, trained from scratch, at 76.27% Dice from one click. Feature 3DGS runs SAM's decoder on 3D scenes. Author-reported. Trade-off: fewer user prompts against training cost and prompt-style sensitivity.ConceptPrototype learningUsually one prototype per class, the mean embedding of its examples; inputs go to the nearest. FedProto shares prototypes, not weights, across federated clients (4 or 41 thousand parameters per round, versus 235 thousand or more). FedProc pulls features toward them. SDC corrects them as the embedding drifts. ProtoSeg clusters several per class for segmentation. Gains are author-reported.ConceptProximal policy optimization (PPO)PPO is a policy-gradient method that clips the new-to-old probability ratio, so one update cannot move the policy far. Twenty-two cards use it as learner, fine-tuner or reward partner, from team learning and evolutionary search to diffusion policies, motion latents, physics rewards, FPGA accelerators and LLM agents.ConceptPsychometric scale validationHow to check that a self-report questionnaire measures what it claims: factor structure (CFA), reliability, convergent and discriminant validity, test-retest stability and method bias. Three author-reported validation studies show the checks and their gaps; two user studies use established scales with no such checks. None reports every check with statistics; method bias is the least covered.ConceptPublish-subscribe messagingPublishers send typed messages to a named topic, and any number of subscribers receive them without knowing each other. This page covers one design, DDS-based topics in ROS 2, with its quality-of-service settings, same-process delivery and security. Only one review card is available so far. Its benchmark is one publisher and one subscriber, so its numbers are indicative only.ConceptQuadratic discriminant analysisQuadratic discriminant analysis classifies with a quadratic boundary between two normal classes that have different covariances. For two normals, the Bayes-optimal rule has this form. The one card so far covers the two-class case, fitting from samples, and computing the error of a quadratic boundary. It does not compare QDA with other classifiers.ConceptQuadratic programmingMinimise a quadratic objective under linear equality and inequality constraints. Legged-robot MPC and whole-body control reduce to QPs solved at high rates, and QPs also appear in teleoperation IK and trajectory smoothing. Solver families trade warm starts, sparsity, accuracy and code generation. A review maps them to problem shape; its ratings are its own judgements.ConceptQualitative coding with LLMsUsing a language model to assign fixed codes from a researcher's codebook to open-ended text (deductive coding). A codebook-style prompt with about one example per code gave the most agreement with experts in one small study, at kappa 0.61 for one dimension and 0.38 for another, both below expert-to-expert agreement. One study so far, on one dataset.ConceptQuality-diversity searchSearching for a whole collection of good solutions that differ in behaviour, instead of one optimum. One card so far: a single loop with a swappable container (grid or epsilon-dominance archive) and selection operator, unifying MAP-Elites and NSLC. Results are simulation-only and author-reported.ConceptQuantization-aware trainingFine-tuning with simulated rounding of weights and activations to low-bit integers, so the network learns to tolerate it. A straight-through gradient makes this trainable, usually with a float master copy. Needed when post-training quantization loses too much accuracy, e.g. per-tensor grids on MobileNets or 4-bit weights. Batch-norm handling differs between sources. Costs a training run.ConceptRandom convolutional kernelsTime series classifiers that convolve each series with many short kernels, pool every response with simple statistics and fit a linear model on the features. MultiRocket uses fixed kernels and is author-reported close to HIVE-COTE 2.0 at a fraction of its training time on 109 UCR datasets. Arsenal uses random kernels in an ensemble and handles multivariate series.ConceptRandom feature approximationRandom feature approximation replaces a kernel with an inner product of fixed random projections. In attention and graph transformers, the sum over pairs can be reordered so cost is linear in size; in Bayesian optimization, it gives cheap posterior function samples. Trade-offs: a feature-size knob sets the accuracy, the approximation is not exact, and trained softmax models do not transfer.ConceptRandom forestAn ensemble of decision trees, each grown on a resampled training set, whose outputs are averaged. Insensitive to feature scaling on independent rows. Repeated measures need changed sampling or a mixed-effects hybrid. Covers subject-level bootstrap, historical summaries, mixed-effects hybrids, multivariate splits, regression forests, and plain forests in applied studies.ConceptRationale extraction (models that show the supporting input)Making an NLP model return a short input span, a rationale, with each prediction. Designs: hard selection, a two-stage pipeline, and top-k soft scores. ERASER finds a BERT pipeline matches human sentence rationales best, but sentence pipelines are unsuited to short spans. LIME or gradient scores are more comprehensive than attention. SciFact applies a pipeline to scientific claims.ConceptRecurrent neural network (RNN)A recurrent network carries a hidden state along a sequence. LSTM and GRU cells recur across the cited work, from forecasting and tracking to OCR, dialogue emotion, text generation and world models. In one author-reported forecasting benchmark, LSTM won the most datasets on best-of-grid accuracy, a ranking that selects on the test split.ConceptRed teamingProbe a model on purpose for harmful replies, score them, then fix the failures. Variants: crowd dialogue attacks, RLHF probing, red-team data for preference training, self-critique, gradient and LLM-driven attacks, multi-turn RL attackers, judge-scored benchmarks, trust tests and system cards. Results are author-reported and not comparable.ConceptReferring segmentationOutput a mask for each object a text query or chat answer names, in one model. PixelLM writes learned codebook tokens and a small decoder turns them into masks, with no separate SAM model. GLaMM writes one segment token per object in a grounded chat, decodes it with a SAM-based decoder and accepts user boxes. Both report gains over LISA, author-reported and mostly on different benchmarks.ConceptReinforcement learning evaluationHow to report deep RL results when each method has only a few training runs. Covers stratified bootstrap intervals, score-distribution profiles, the interquartile mean, optimality gap and probability of improvement. Most choices depend on runs per task and on whether a headline number or a comparison is needed. Results are author-reported.ConceptReinforcement learning from AI feedback (RLAIF)Replaces human harmlessness labels with AI preference labels. A feedback model picks the less harmful of two answers under a written principle, a preference model is trained on those picks, and RL optimises the policy against it. Helpfulness can keep human labels. Evidence is one paper, author-reported, mostly plots.ConceptReinforcement learning from human feedback (RLHF)Post-training that fits a reward model to human preference judgments, then optimises the model with RL after supervised fine-tuning. DPO and KTO drop the reward model and RL loop; LIMA drops RLHF; Constitutional AI swaps human harm labels for AI feedback. Extensions cover data, images and video; sycophancy and red-team studies test side effects. Results are author-reported.ConceptReinforcement learning with verifiable rewards (RLVR)Fine-tuning by sampling answers and scoring each with a rule that can be checked automatically (IoU, class match, boxed maths answer, solver objective, kernel speed, simulated lap error, final answer of a worker chain, success in a 3D game task), then favouring answers that do better than their peers. Author-reported gains. Needs a checkable output.ConceptRelation networkA set-scoring network that applies one learned function to every pair of object embeddings, sums the results and scores the sum with a second network. Variants differ in what feeds the pairs: a CNN trained from scratch, a frozen pretrained encoder, ground-truth factors, or unsupervised disentangled encoders. One cited paper, on Raven-style puzzles.ConceptRelational graph convolutionA graph layer that gives each edge type its own weight matrix, so a node's neighbours are transformed according to how they link to it. Use it when edges fall into a few meaningful types and the whole graph is available up front. The trade-off: the number of weight matrices grows with the number of types, and the one card here leaves several layer details unstated.ConceptRelative positional encodingPosition information in attention that depends on the offset between query and key, not on an absolute index added to the input. Swin learns a bias per window offset, and Swin V2 replaces the table with a log-spaced network. Adding a 2D bucketed term to DeiT-S's absolute encodings gains 1.5 top-1 (iRPE); removing ConveRT's bias costs 0.4 R100@1.ConceptReplaced token detectionTrain a small generator to fill masked slots, then train a discriminator to label every token as original or replaced. The discriminator sees all tokens, not only the masked ones. The main design choice is how the generator and discriminator share token embeddings: naive sharing conflicts, and gradient-disentangled sharing is the variant reported so far.ConceptResidual learningA residual unit computes a learned function F(x) and combines it with a shortcut of its input, so the block learns a correction to x. Thirteen papers use it across vision, spiking networks, forecasting, offline RL and encrypted inference, with different blocks and combine rules. Results are author-reported.ConceptRetrieval-augmented generationGives a language model passages fetched at answer time, so its answer rests on material outside its training. Variants differ in how they find passages, from top-k embeddings to keyword counts or a model that decides when to retrieve. Often beats fine-tuning on the same documents, author-reported. Trade-off: answers depend on retrieval, thresholds need calibrating, and measurements are thin.ConceptReturn-conditioned sequence modelingConditioning a policy on a target return: a transformer predicts return, observation, action and reward tokens and samples a high target return at play time; a diffusion policy can take a return-to-go embedding instead. Two papers, both with author-reported results. The main design choices are how often the return appears, how it is sampled and how it is defined.ConceptReverse attentionReverse attention turns a coarser prediction into 1 minus its sigmoid, a map of what that prediction has not yet explained, and uses it to guide the next refinement. Used in camouflaged object detection, either as one reverse step on a fixation map or as a coarse-to-fine cascade of group-reversal blocks. The gain in SINet-V2's own ablation is small (CAMO S-measure 0.814 without it, 0.820 with it).ConceptReview-based recommendationPredicts a user's rating of an item from the text of reviews, not only from past ratings. DeepCoNN, the one card so far, runs a CNN over all reviews by the user and all reviews of the item, then scores the pair with a factorization machine. It reports 7.6% to 8.7% lower MSE than the best baseline on Yelp, Amazon and Beer, with the largest gains where ratings are scarce.ConceptReward designChoosing the terms of a reinforcement learning reward and where they come from. Options: hand-set terms with a weighted sum or a priority order, physics-based shaping, potential-based shaping from a task graph, learned rewards from demonstrations, a vision-language model or a world model, and redistribution of a sparse team return. Results are author-reported and not on a common benchmark.ConceptReward machineA finite-state machine over task propositions whose transitions carry rewards, so an agent can track sub-goal progress. The usual setting uses an expert-written machine with known propositions. One paper infers the machine and its propositions from unlabelled RGB demonstrations of block tasks, then trains a DQN with potential-based shaping. Results are author-reported and simulation-only.ConceptReward modelingLearning a score from judgements of outputs or from a sparse return, then using it to train an agent or generator, or to pick among candidates. Covers a user-in-the-loop agenda, an Atari agent, two image scorers, a pairwise LLM ranker, a group-consensus reranker and a learned split of team returns. Results are author-reported.ConceptReward-proportional samplingGenerate discrete objects step by step so that each finished object is sampled with probability proportional to a positive reward, not only the single best one. Core issue: many build orders lead to one object. Variants: flow matching and trajectory balance (GFlowNets). Trade-off: flow training and tuning against simpler return-maximising RL.ConceptRobot simulation for learningTraining and testing robot policies in a simulated world before real hardware. Orbit and Isaac Gym run many physics copies in parallel on one GPU. Replica supplies indoor scenes, and many papers instead score policies on simulated benchmarks such as LIBERO, RoboTwin or RLBench. Choose a simulator by rendering needs, deformable objects and GPU budget.ConceptRobotic grasp generationPredicting where a robot gripper should close on an object, and how it should approach and open. Nine papers: sample-and-classify, per-point contact, latent diffusion, Gaussian-scene grasping, dexterous RL priors and geometric grasp points. Author-reported, on different setups, so the numbers are not comparable.ConceptRotation-equivariant CNNA CNN backbone whose feature maps keep one block per discrete rotation, so a rotated input gives a rotated feature map and the weights are shared across orientations. ReDet uses it for oriented aerial detection, with a 12 Mb backbone against 103 Mb and higher mAP on DOTA-v1.5, plus an orientation-aware RoI pooling step. One paper so far.ConceptSafe screeningDrops features of a Lasso problem that are provably zero before solving, using a ball around the dual solution at a nearby penalty. Sequential EDPP is the tightest form in the paper; it needs the previous path solution and assumes it is exact. Author-reported speedups of 30 to 230 times on high-dimensional data. One source so far.ConceptSafety filterA safety filter overrides a nominal controller only when needed to keep the system in a safe set. Backup-based filters are compared in a 2026 review, which finds gatekeeper keeps the nominal controller most often, in simulation only. A learned value-function filter is tested on a quadruped. A driving model's speed-capping layer is also covered. The term also covers diffusion content filters.ConceptSalient object detectionSalient object detection predicts a per-pixel map of the most eye-catching object, scored against a binary mask. S-measure scores structure where AP, AUC and Fbw only count pixels. EGNet adds an edge branch (max F 0.893 on DUTS-TE). Variants add depth (D3Net, SwinNet), thermal (SwinNet), camouflage (SINet-V2, ANet, COD-Rank), eye fixations (DVA) and video (no optical flow).ConceptScene graph generationDetect the objects in an image and predict a relation for each pair: a subject-predicate-object triplet, or a blocking dependency for bin picking. Image methods fall into CRF, TransE, CNN, RNN/LSTM, GNN, one-stage message-passing and query-graph families. Predicates are long-tailed, so compare mean recall per predicate. Robot papers use their own tasks and cannot be ranked against VG150.ConceptScore distillation samplingScore distillation sampling (SDS) turns a frozen diffusion model into a training signal for a differentiable generator, such as a 3D scene or a one-step student. The core idea is the noise-prediction residual, sent back through the generator. The trade-offs are a noisy gradient that needs weighting and regularization, and a prior that must be queried at every step.ConceptSecret sharingSecret sharing splits a value into parts that each party holds, so no single party learns it and the parties can compute on it together. Only one card so far: a four-party framework with at most one corrupt party, using additive, replicated and masked sharings over a 64-bit ring. Sharing types and conversions are described, with author-reported speed-ups against ABY3.ConceptSecure multi-party computationParties compute a function over their private inputs and learn only the output. Secret sharing splits each value across parties, and the protocol works on the shares. Only one card so far: a four-party framework, at most one corrupt, for ML training and prediction. Its author-reported speed-ups over a three-party baseline come from the authors' own re-implementation.ConceptSeed-variance evaluationHow to tell whether a training-based score difference is real: report score distributions over random seeds rather than one run. Three variants: a seed distribution for one fixed setup, a seed-pair noise floor across random configurations, and a paired comparison of one design choice. The evidence comes from one paper on LSTM sequence tagging, and the official code does not fix seeds.ConceptSegmentation evaluation (scoring foreground, saliency and segmentation masks)How to score a predicted mask, box or saliency map against ground truth. Pixel-count measures (IoU, Dice, F-beta, pixel AP, AUC, Fbw) ignore object shape; S-measure also rewards intact structure. Detection uses AP over box IoU thresholds. A consensus framework maps the problem to metrics. Trade-off: comparability with prior work versus agreement with human and downstream judgements.ConceptSelf-attentionSelf-attention lets every token in one sequence weigh the other tokens of that sequence when building its new representation. Variants differ in who may attend to whom (full, windowed, sparse, linear, masked, shared, memory-backed or gated), and that choice sets memory cost, data needs and how easily a pretrained model can be extended.ConceptSelf-consistencyReplace one greedy chain-of-thought decode with many sampled chains and return the most common final answer. On PaLM-540B, 40 samples lift GSM8K from 56.5% to 74.4% (author-reported). It needs short answers that can be compared for equality, and about 40 times the decoding cost. Gains are smaller on smaller models.ConceptSelf-playTraining an agent by playing copies of itself, which can yield a strong partner for those copies but not necessarily for a stranger. Covers plain self-play, other-play (random symmetry relabelling), search-guided self-play (AlphaGo Zero style) and self-play as a curriculum source. Three papers; results are author-reported.ConceptSelf-refinement (draft, critique, rewrite loop)A model drafts an answer, critiques it, rewrites it, and repeats for a few rounds. The critique can come from the model alone, from running the output, or from a ranking among candidates. At test time no weights change. Gains are largest when the critique is grounded in a run or a test. Unaided self-checks on math and multiple-choice reasoning are mixed. Needs a strong instruction follower.ConceptSelf-supervised learningPre-train an encoder on unlabelled data with a pretext task whose targets come from the data itself: rebuild hidden parts (generative), pull matching views together and push others apart (contrastive), or train a generator against a discriminator. Pick the family by data modality, batch and memory budget, and whether the features will be fine-tuned or probed linearly.ConceptSelf-trainingTeacher-student loop: a model pseudo-labels unlabelled inputs and a student, often trained on stronger augmentation, learns from the kept labels. Variants differ in which labels they keep (confidence or entropy cut-offs, or soft labels) and in the teacher. Trade-off: extra passes and a second model versus gains from unlabelled data, which can also hurt.ConceptSemantic occupancy mappingKeeps a 3D memory of which space is occupied, free or unknown, with a semantic class for occupied points, updated as posed views arrive. One card so far: GEM-Occ turns RGB predictions into semantic Gaussians plus free-space rays, fused into submaps. Author-reported on HIOcc: 61.37 local IoU, 0.81 MB/m² at building scale. Code not yet released.ConceptSemantic parsingTurning a question or problem statement into a formal program that a database or solver runs. TaBERT, a table-pretrained BERT encoder, reaches an author-reported 51.8 WikiTableQuestions test mean (K = 3). Logic-LM, where an LLM writes a logic program and a solver runs it, beats GPT-4 chain-of-thought on four of five datasets. Pretraining cost and non-commercial code limit reuse.ConceptSemantic scene completionPredict a dense 3D grid in which every cell is empty, a semantic class, or unknown, including cells the sensor never saw. Covers a LiDAR variant that trains completion as an auxiliary task and drops it at inference, and an online camera variant that fuses posed RGB into a persistent occupancy memory. Gives the trade-offs each reports, and warns that the two papers' numbers are not comparable.ConceptSemantic segmentationAssign a class label to every pixel. Covers ViT encoders with light decoders, label-free, semi-supervised, domain-adaptive, continual, weakly and text-supervised, few-shot, in-context, promptable, open-vocabulary and diffusion-based variants, and domain-specific models (3D medical volumes, polyps, aerial and satellite imagery, driving, SLAM), with the trade-offs each one reports.ConceptSemi-supervised learningTraining on a few labels plus many unlabelled inputs. Families: generative, consistency, graph, pseudo-label and hybrid methods; variants for dense prediction (images, LiDAR), detection, clustering, text, tabular and sensor data. Trade-off: extra passes and stored state; unlabelled data can hurt.ConceptSentiment analysisSentiment analysis labels the opinion in a text, as polarity, a score or emotion. Lexicon tools, CNN and LSTM classifiers, fine-tuned pretrained encoders, multimodal fusion and prompted LLMs are the options. Reported numbers use different datasets, labels and metrics and cannot be compared; sarcasm, emoticons and aspect-level opinions are hard cases.ConceptSequence labelingGive each token in a sentence one label, such as a span boundary or an entity type. Covered: fine-tuned encoders with task heads (English, Dutch and scientific BERT), BiLSTM taggers with softmax or CRF output, per-word reading features in a BiLSTM-CRF, cascaded tagging, a spaCy pipeline, data augmentation, few-shot prompting, prefix tuning and seed-distribution evaluation.ConceptSeries decompositionSplits a series into parts modelled separately: a moving-average trend and remainder (DLinear), learned trend and seasonal features (CoST), or level, growth and season peeled off inside a network (ETSformer). DLinear beats plain Linear in 21 of 36 Table 2 cells. DLinear's MSE is lower than ETSformer's in all 24 rows of ETSformer's table. Author-reported, mostly single runs.ConceptShape-prior regularizationPenalise a network's outputs for being implausible by comparing low-dimensional codes of predicted and true masks or images, learned by an autoencoder or predictor. The term is used only during training, so inference cost is unchanged. Variants differ in output type (label maps or intensity images). Choose by what the network outputs and whether you can afford a two-stage training schedule.ConceptShapley valueSplits a shared score among players by averaging each player's marginal contribution over every order in which they could join. Covers exact enumeration, permutation and regression sampling, and Monte-Carlo credit for steps in LLM-agent runs, with the cost, axiom and sign trade-offs. Sources are one survey with no experiments and one paper validated only on synthetic models.ConceptShared autonomyA robot and a human both shape one command, and the robot's share is set by a fixed weight or a takeover rule. Compares a blend weight, tested in a 24-person user study of arm tracking, with a learned-motion-primitive assist that takes over a grasp on a humanoid. Covers the constraints each needs and what the results support. Both papers are author-reported with small samples.ConceptShifted-window attentionShifted-window attention limits self-attention to non-overlapping windows and moves the window grid every other block, so tokens can reach across window edges. Video Swin uses 3D windows, UniMatch uses 2D windows at 1/8 resolution, and GRL pairs Swin V2-style windows with anchored stripes. Ablation gains are author-reported, without run-to-run variance.ConceptSiamese networkOne encoder with shared weights reads two inputs, then they are compared. Variants: cosine similarity between stored sentence vectors, learned per-scale differences for change masks, or one joint ViT over both inputs that drops the separate branches. Pick by whether you need a reusable vector or a dense output, and by labels and compute.ConceptSignal temporal logicHow to write timed rules over real-valued signals and turn them into a controller cost. Robustness degree gives a signed margin, and a smooth encoding makes it solvable inside a nonlinear MPC. One card so far, author-reported, on bipedal push recovery.ConceptSigned distance fieldA grid that stores, for each cell, the distance to the nearest surface, with a sign for inside versus outside. Common pipelines fuse depth into a truncated SDF (TSDF), then derive a Euclidean SDF (ESDF) at a chosen resolution. Covered so far by one paper: a GPU depth-to-ESDF stack for high-DoF robot motion generation, with author-reported speed and memory results.ConceptSim-to-real transferMoving a model trained on synthetic data onto a robot or real images. Twenty-four sources, from humanoid and quadruped locomotion and grasping to mobile manipulation, drone payloads and vision-model testing. Author-reported.ConceptSimultaneous localization and mapping (SLAM)SLAM estimates where a robot or camera is while it maps its surroundings. Sixteen cards cover lidar, visual, lidar-visual-inertial, monocular and RGB-D systems, scene-graph and open-set semantic mapping, object-level and neural-SDF mapping, two datasets, three surveys and a navigation paper. Numbers are author-reported and rarely comparable.ConceptSkip-gram with negative sampling (SGNS)Learns vectors by pulling observed (item, context) pairs together and pushing b randomly drawn negative items away. Skip-gram applies it to random-walk (node, feature) pairs, where dot products approximate shifted PMI. Sent2Vec reverses the direction, predicting each word from its sentence, which gives sentence vectors. Both are author-reported, and their settings differ.ConceptSliding-window attentionEach position attends only to a fixed window of nearby positions, so the cost per position does not grow with sequence length. Mistral 7B uses a causal window of 4096 with a rolling-buffer cache, Gemma 2 alternates local and global layers, and ActionFormer uses a two-sided window over video steps. Author-reported.ConceptSliding-window estimationSliding-window estimation re-solves a state estimate over only the last few time steps, and folds older steps into a prior. It sits between a live filter and a batch smoother. One card so far: a continuum-robot shape estimator that is close to batch accuracy at about 10 ms per step, with a delay of up to the window length.ConceptSoft prompt tuningReplace hand-written prompt tokens with trainable continuous vectors, usually while the model stays frozen, and fit them by gradient descent. Variants differ in where the vectors go, what they are fitted on, and whether several prompts are mixed. Reported gains are author-reported, on different tasks, and not comparable.ConceptSparse Bayesian learningSparsity from a Bayesian prior: each weight gets a zero-mean Gaussian whose variance is learned, and weights whose variance collapses are pruned. Used for neural architecture search and structural pruning. Trade-offs: the search caches feature maps in memory, and the only known source is one author-reported paper with unresolved inconsistencies.ConceptSparse autoencoderA sparse autoencoder learns sparsely active features from one layer of a language model. Use an unsupervised one to find features in your own model, a supervised one to bind named concepts to fixed slots, or a released labelled one as a document embedding. Reconstruction is lossy, and most results are author-reported on narrow setups.ConceptSparse distributed representations (SDRs)Encode each input as a small set of active bits out of many, here 40 of 2048 (about 2%), so a sequence memory can match inputs by shared active bits. Variants use random codes per symbol or pool several encoded fields into one code, and read the code back by overlap or a trained softmax. Trade-off: only this one sparsity level is tested, and the multi-field encoder sizes are not stated.ConceptSparse fine-tuningAdapt a pretrained model by updating only a fixed subset of its existing weights and freezing the rest. The subset can be chosen randomly, by rule, by projection or by gradient scores; SAM ranks entries by squared gradients from burn-in steps and reports the best average on seven RoBERTa-base tasks at 0.5% tunable weights (author-reported).ConceptSpectral clusteringClustering by the eigenvectors of a graph Laplacian built from pairwise affinities; the classic route needs an N-by-N affinity matrix and an O(N³) eigendecomposition. Anchor-based transfer cut solves a p-by-p eigenproblem instead. A 270-dataset R benchmark ranks spectral first at default settings, with K set to the true class count.ConceptSpectral methodsEstimate hidden structure from the leading eigen- or singular vectors of a matrix built from data. A monograph gives perturbation bounds for denoising, PCA, SBM clustering, ranking, phase retrieval and matrix and tensor completion. Also covered: spectral matching for point-cloud outliers, spectral initialisation of one-hidden-layer nets, and signed, directed graph filters.ConceptSpectral normalizationCapping the spectral norm (largest singular value) of weight matrices after each update, so a network cannot squeeze distinct inputs together. The cap is a hyperparameter found by search, and the one-step power estimate is approximate. Cards: SNGP, one half of a single-model uncertainty method; Gouk et al., an l2 cap as a regulariser; RL-GAN-Net, a setting in a latent GAN.ConceptSpectrogramA time-frequency grid of signal strength, computed from short-time Fourier transform frames. Log-mel is the common audio variant, used as classifier input, generation target, VAE input or robot-policy input. Papers often leave the FFT, hop and bin settings unstated, and the released settings are not interchangeable.ConceptSpeculative decodingCuts per-token latency of memory-bound LLM decoding: a cheap drafter proposes several tokens, the target model checks them all in one forward pass, and an accept/resample rule keeps the target's output distribution. Drafters range from a small separate model to target-side heads or one parallel block. The cost is extra compute and memory, so the gains shrink as batch size grows.ConceptSpiking neural networksSpiking neural networks send sparse on-off spikes through leaky neurons. Training is hard because a spike has no useful gradient. Options: surrogate-gradient BPTT, with optional residual blocks for depth and optional attention on the membrane potential; ANN-to-SNN conversion; spike-time gradients; online learning; learned delays. Speech benchmarks: SHD and SSC. Results are author-reported.ConceptStacked generalizationStacked generalization trains a meta-model on several base models' outputs. A rice-image stack reports 100% on its own set, and a card-fraud stack adds 0.008 AUC over logistic regression. A class-imbalance benchmark finds stacks best on four of 16 sets. All results are author-reported.ConceptState-space model (SSM)A state-space model carries a fixed-size state through a sequence, so cost grows linearly with length. Covered: selective scans (Mamba), diagonal S5, 2D image scans, hybrid backbones, scalar-decay SSD (Mamba-2), video, LiDAR occupancy (OccMamba) and robot control (LocoMamba). xLSTM is a recurrent neighbour, not an SSM. Numbers are author-reported.ConceptStick-slip frictionFriction that alternates between sticking and sliding, so a small oscillating force can move an object in steps. Vibration on a parallel gripper uses it to rotate a thin object (author-reported: 1.5 vs 3.8 degrees orientation error with duty-cycle vs constant vibration). Main trade-off: the switching threshold depends on friction values that papers often do not state.ConceptStochastic block modelA latent-group graph model: each node has a hidden group, and edge probability depends only on the pair of groups. A review covers degree-corrected, valued, mixed-membership, overlapping, dynamic, text and nested variants, plus fitting and choosing the number of groups. Two GNN papers use a contextual and a signed-directed variant as synthetic benchmarks.ConceptStructural similarity (SSIM-style comparison of maps)SSIM-style measures compare mean (luminance), spread (contrast) and correlation (structure). S-measure applies a constant-free SSIM per block around the object centroid. A no-reference SR metric uses a structure-only correlation between image bands. Explanation maps can be compared with SSIM after min-max scaling, but high SSIM can coexist with weak rank agreement.ConceptSubword tokenizationSplit words into learned pieces smaller than words, so a fixed-vocabulary model can still read rare and unseen words. Covers BPE and joint BPE, character bigrams, and vocabularies for Spanish, Dutch, Arabic and science, plus byte-level input with a learned tokenizer. Gains are author-reported.ConceptSuccessive halvingBudget allocation that starts many candidates on a small budget, keeps the best 1/eta each round and multiplies the survivors' budget by eta. Covers its use in Auto-sklearn 2.0, its pairing with Bayesian optimization, and when it did or did not beat full-budget training. Author-reported, one paper.ConceptSupport vector machineA classifier in several roles: an RBF SVM on fixed features (97.35% on CK+), a linear SVC on TF-IDF spam (0.991 accuracy), a linear SVM on EHR features (AUC up to 89.8%), an SVM on terrain sensor statistics (95.87%), an SVR baseline, and an active-learning query model. Theory: gradient descent on linear networks reaches the hard-margin SVM. Numbers are author-reported.ConceptSurface normal estimationPredicting a unit surface normal for every pixel of an image, or a normal rendered from a 3D model. Normal labels are scarce, so the main choices are whether you have normal labels or only depth labels, whether normals are predicted jointly with depth, and whether a pretrained normal model can supervise them. Two papers are covered.ConceptSurfel mappingSurfel mapping merges each new observation into a map of oriented points or discs. Covers SuMa and SuMa++, which align LiDAR scans to surfels and use labels to drop moving-object surfels (KITTI, author-reported), and ConceptFusion, which fuses language-aligned features into a posed point map with no alignment step.ConceptSurrogate gradientTrains spiking neural networks with backpropagation by replacing the zero or infinite derivative of the spike threshold with a smooth stand-in on the backward pass only. The forward pass keeps the hard spike. The trade-offs are memory for the stored time steps and no update for neurons that never fire.ConceptSurrogate modelingA learned stand-in for an expensive simulator or physical system, so design sweeps and gradient estimates run cheaply. Variants differ in what they predict: one objective, a field, per-node stress, a material law with built-in thermodynamics, a gradient-only digital model, a time series of modal amplitudes, or a score for finished runs that saves no simulator calls.ConceptSurvival analysisModelling the time until an event, such as churn, death or machine failure, when some records are censored. Covers the risk-model, loss and time-dependence choices, the sampled-control training trick, and the metrics used to compare models. Three cards so far; all reported results are author-reported, and they are not comparable across papers.ConceptSycophancyAn assistant shifts its answer, praise or verdict toward what the user says they believe or like, even when the first answer was right. Four behavioural probes measure it, and preference data can reward it. One paper so far; all results are author-reported on five 2023 assistants.ConceptSymbolic aggregate approximationTurns a numeric series into short words over a small alphabet by averaging segments and binning them at Gaussian quantiles. Covers classic SAX, windowed SAX with dilation and stride, and a bag-of-words classifier that is deterministic and explainable. Author-reported: accuracy tied with ROCKET, but not the most accurate method.ConceptSynthetic query generationTrains a retriever or embedding model on text an LLM writes, instead of labelled search data. Promptagator writes one query per document and keeps round-trip-consistent pairs for a small dual encoder. E5-Mistral has an LLM invent task-specific examples in 93 languages for a 7B model. Results are author-reported and cover different benchmarks.ConceptSystematic generalizationHandling new combinations of parts seen in training. Position papers argue for dynamic binding into reusable object representations, or for causal, verbalisable variables with sparse changes. CREPE reports lower retrieval on unseen compounds. GrailQA's BERT ranker reaches 53.9 F1 on compositional questions (author-reported).ConceptTable question answeringAnswering a question from a table by selecting cells, generating text, running a program over the table, chaining table tools chosen by a planner, or encoding the table for a parser. Four papers cover six routes, from fine-tuned chart and text models to training-free program writing. Gold tables score far above extracted ones, so the table source matters most. All numbers are author-reported.ConceptTabular foundation modelA tabular foundation model is pretrained on synthetic tables and used as a frozen regressor on a new table, with no training step. Forecasting can use it by turning time into table features. One card so far: about 11M parameters, competitive zero-shot forecasts, strongest with known-future covariates, no trend extrapolation, and much slower per series than the compared baselines.ConceptTask and motion planningChecking a language model's robot plan before the robot moves. Text2Motion has an LLM propose skill sequences over learned Pick, Place, Pull and Push skills, then scores each with geometric feasibility planning and out-of-distribution rejection. Shooting, greedy and hybrid search are compared. Author-reported: 82% vs 13% for prior language planners on six simulated tasks.ConceptTask decompositionSplitting a task into sub-tasks, each solved by a prompt, function or policy. Decomposed Prompting has an LLM write the program; DIN-SQL runs a fixed prompt pipeline for text-to-SQL; SMART-LLM stages prompts to plan code for a robot team; MEM keeps a text memory; CycleVLA retries subtasks after a VLM check. All results author-reported. MEM and CycleVLA need trained robot policies.ConceptTeacher-student learningA fixed teacher model labels the data and a same-shape student is trained on it. Question: can the student recover the teacher's weights, not only its predictions? For one-hidden-layer nets with Gaussian inputs and noiseless labels, tensor initialisation plus gradient descent does so with samples roughly linear in input dimension. Trade-off: strong assumptions, theory only.ConceptTeleoperationA human drives a robot in real time, most often to record demonstrations or to take over a running policy. Covers how operator motion is sensed (cameras, gloves, leader arms, VR trackers), mapped to robot joints, blended with autonomy and fed back as force, with the cost and evidence of each. All evidence is author-reported on different systems, so the numbers do not rank against each other.ConceptTemperature scalingDivides a softmax model's logits by one learned positive number after training, without changing the top class. It is the one-parameter case of Dirichlet calibration's linear family. On average ranks it trails Dirichlet ODIR on deep networks and Dirichlet L2 on non-neural classifiers. LogitNorm's training-time temperature is a different use. Sources are author-reported.ConceptTemporal convolutional network (TCN)Dilated 1D convolutions over time in residual blocks, or gated and causal in a survey. Seen in a forecaster, a skeleton recognizer, a traffic graph model, an EMG regressor and a forecasting library, and described in two GNN surveys. Receptive field is set by layers, kernel and dilation. The papers measure different tasks, so their numbers are not comparable.ConceptTemporal-logic runtime monitoringHow to check that a run obeys rules about order and duration, not only its end state. Write each rule as a temporal-logic formula over per-step predicates, compile it to an automaton, step it through the trace and score violations per triggered event. Two cards: a robot-manipulation benchmark with author-reported results, and a position paper with no experiments.ConceptTensor decomposition (for parameter-efficient adaptation)Stack the same-shaped weight matrices of a frozen network into one tensor, then train a low-rank factorisation of its increment, with factors shared across layers. FacT (Tensor-Train or Tucker) matches NOAH on VTAB-1K with about a fifth of its trainable parameters, author-reported. Trade-off: weights must share a shape, and results use test-set selection.ConceptTest-time augmentationTest-time augmentation (TTA) runs a classifier on transformed copies of one input and combines the outputs. Plain averaging can flip correct predictions to wrong ones. Learned non-negative weights per view, or per view and class, can help when a labelled held-out split exists. Random-view averaging is also used in CT lesion detection. Author-reported gains are mixed across settings.ConceptTest-time scalingSpending more inference compute per question, either by letting a model write a longer reasoning trace or by sampling several answers and voting. On an edge GPU, longer answers dominate latency and energy, and a 14B reasoning model reached 80.6% on MMLU-Redux at 259 s per question (author-reported). Budget caps and parallel voting trade accuracy against time.ConceptTest-time trainingAdapt a model to each test input before it predicts, using a loss that needs no label, then discard the adapted state. Two papers so far: masked-patch reconstruction of a ViT (20 SGD steps, ImageNet-C level 5) and entropy-based prompt tuning of CLIP over diffusion-generated views (4 steps). Gains are author-reported. Trade-off: robustness to unseen shifts against much higher per-input cost.ConceptText classificationAssigning one label to a text or sentence pair. Narrow tasks: linear models on TF-IDF, FastText and CNNs stay strong. Few labels: a contrastively tuned sentence encoder with a logistic-regression head nears prompt methods. Cross-domain transfer is weak, and dataset artifacts or altered input distort scores. Zero-shot LLM labels vary widely by task and need checks.ConceptText de-identificationRemoving details that identify a person (names, dates, places, IDs, contact data) from text before it is shared. One paper so far: prompt-only redaction of clinical notes with ChatGPT and GPT-4. Author-reported entity accuracy on the i2b2/UTHealth set is 0.929 for ChatGPT and 0.99 for GPT-4, the latter on 50 hand-run samples.ConceptText embeddingFixed-size vectors for text or code, so similar inputs sit close. Variants differ by training signal: weak pairs, denoising, silver labels, teacher distillation, few-shot tuning, n-gram averaging, LLM-written data or an unlabelled LLM. Frozen embeddings can feed a classical classifier when labels are scarce. Sparse concept latents are interpretable. Pick by labels, compute and serving budget.ConceptText simplificationRewriting dense text so a reader without specialist training can follow it. Two papers so far. In a radiology study, 15 radiologists rated 45 chatbot rewrites: mostly correct, but with errors and dropped results. A service robot restates answers in German easy language through an external API, with no measurements.ConceptText summarizationShorten a text, such as news, spoken dialogue or clinical notes, into a summary that keeps the key content. Extractive methods copy sentences, abstractive methods write new ones, and hybrid methods do both. Fine-tuning needs labelled pairs; a prompted LLM needs none. Long inputs need a long-context model. Automatic scores can disagree with human ratings.ConceptText watermarkingHides a statistical bias in the word choices of generated text so a detector can later find it with a simple test. Red-green watermarks need the preceding context, which diffusion language models often lack. One paper so far adapts the method by tilting each denoising step's logits with the expected green-count gradient. Author-reported: 99% TPR at 1% FPR on LLaDA and Dream.ConceptText-guided image editingChanges an image from a text prompt. Covered: training-free Stable Diffusion editing that keeps layout; CLIP-driven optimisation (VQGAN, CLIPstyler); per-input layered editing (Text2LIVE); CLIP-steered StyleGAN2 in three routes; a trained outfit GAN; subject-driven and per-image Stable Diffusion. Trade-offs: compute per image, what each keeps, training.ConceptText-to-3D generationMaking a 3D object from a text prompt by optimising a 3D representation until pretrained models rate its renders as matching the text. Two variants: CLIP-Mesh (textured mesh and normal map, CLIP and a diffusion prior, zero-shot) and GSGEN (3D Gaussians, Stable Diffusion SDS plus a Point-E prior, about 40 minutes per asset). Results are author-reported.ConceptText-to-image generationGenerating images from captions. Nine generators in eight families, from GANs and token transformers to latent diffusion, adapters, layout steering, one-step distillation and a multimodal model. Also a safety steer, a reward-tuned generator, a watermark, a dataset, three scorers and two evaluators. Results are author-reported; compare within a row.ConceptText-to-image personalizationTeaching a pretrained text-to-image diffusion model a new subject, such as a pet, an object, a person or a style, from a few photos or one image. Covers cross-attention key and value fine-tuning, joint and merged multi-concept training, pre-trained subject encoders, tuning-free multi-person generation, single-image editing, and training-free style sharing. Six compiled cards.ConceptTextual inversionLearning one new token embedding for a frozen text-to-image diffusion model, so that a made-up word stands for one object or concept. Covers the single-image variant from RealFusion: augmented copies of one image, batch, learning rate, token initialisation, and where the released code departs from the paper. One compiled card so far.ConceptTexture analysis and classificationDescribes an image region by its texture so that a classifier can tell texture classes apart. Descriptor families are statistical, structural, model-based and transform-based, and combined methods stack them. The one cited source is a 2019 survey with no experiments of its own, so its accuracy numbers come from other papers and cannot be ranked.ConceptTheory of mindTheory of mind (ToM) lets an agent infer what other agents believe, intend and feel. A survey sorts the options into two paradigms: prompting an LLM to reason about others, and explicit model-based inverse planning. Prompting is light and black-box; model-based inference is interpretable and costly. Benchmarks are mostly text-only and offline. Results are author-reported.ConceptThin-plate spline (TPS) warpA smooth 2-D mapping or surface built from point pairs or samples, with an affine part plus radial basis terms. An exact TPS warp moves image content for animation and gives lower keypoint error than affine motion, with slightly worse identity error. A smoothing TPS, with cross-validation, splits photogrammetry displacement error into systematic and random parts.ConceptThompson samplingPosterior sampling: keep a posterior over each option's value, draw one plausible sample, act on it, then update. Three papers use it: to find the best and worst of 320 prompt formats within a call budget, to pick queries from a mixture of Gaussian processes, and to search a trust region under a safety threshold. Trade-offs: the prior or surrogate, and results that depend on setup.ConceptTime series classificationGives one label to a whole series. Compares a GPS CNN, MultiRocket, HIVE-COTE 2.0, contrastive pre-training (ECG; TS-TCC), Transformers on clinical and satellite series, BORF's symbolic counts, image encodings, robot classifiers, a Random Forest for burned areas, a tutorial and a benchmark. Author-reported results use different protocols and cannot be ranked against each other.ConceptTime series forecastingPredicts future values of a series, many linked series, a weather grid or an agent's path from its past. Forecasts also pretrain on clinical records and score anomalies. Options range from MLPs and recurrent cells to diffusion, graph, neural-ODE and foundation models. Most evidence is author-reported, and the cards' numbers are not comparable.ConceptTime-optimal controlLinear minimum-time control: pick inputs that reach a goal state at the lowest cost of effort plus a weighted duration. Closed forms exist for velocity and acceleration control. The paper uses them as a heuristic for A* over a lattice of motion primitives, and reports a large node reduction for one start state. Only one source so far; no head-to-head comparison.ConceptTime-series imputationFilling missing values in a time series. Two methods are on the wiki. GRIL, a bidirectional graph-recurrent imputer, suits multivariate series whose channels sit on a fixed graph and beats BRITS on the paper's real data. A denoising autoencoder on Gramian angular field images fills gaps in single series, with gains only on the missing points. A 2023 survey sets out the task and its settings.ConceptTimed elastic bandLocal trajectory planner that optimises a band of poses between robot and goal, with time intervals and penalty terms for limit violations. Two extensions: MP-TEB adds predicted-human social costs (author-reported best total time in 3 of 4 simulated scenes); RTEB adds a hybrid A* recovery layer for infeasible trajectories (lower distance and time in dense obstacle tests).ConceptToken mergingCutting the token count of a transformer by averaging the most similar tokens inside each block, instead of deleting them. Token Merging (ToMe) matches tokens bipartitely on attention keys and merges a fixed number r per layer, with no added parameters. Off the shelf, ViT-L/16 MAE runs about 2x faster for 1.7 points less top-1 (author-reported).ConceptToken pruningRemoving tokens from the sequence so later computation runs over fewer of them. FastV prunes vision-language image tokens after an early layer, ranked by attention, with no training. ToMe merges similar vision-transformer tokens in every block, a close relative. HERMES evicts KV-cache video tokens by layer role while a stream arrives. Results are author-reported and not comparable across methods.ConceptTool useLet a language model emit a call to an external tool, run it outside the model, and feed the result back into the context. Twenty-seven wikillm cards cover prompt-only, few-shot, planned, structured-loop, learned, trained, gated, guarded, versioned and API-writing variants. Nineteen report measured results, most mixing tool use with other changes.ConceptTopic modellingFinding the main topics in a text collection with non-probabilistic (LSA, NMF) or probabilistic (PLSA, LDA) models, and with extensions that put text on graph edges or nodes. Plain topic models have no hyperparameters or results on these cards; the graph variants report author-reported numbers.ConceptTopological map navigationMoves an agent to places on a graph of places it has seen. LM-Nav matches route landmarks to nodes and finds the cheapest walk visiting them in order. SAME learns a policy over the graph. NoMaD samples local moves from a diffusion policy, using the graph for exploration. Trade-offs: LM-Nav needs a map and recognisable landmarks; SAME needs training data; NoMaD takes photo goals, not text.ConceptToxicity detectionLabelling a text as toxic or not. Detectors trained on social-media or other public text transfer poorly to real chatbot prompts: on ToxicChat, an in-domain RoBERTa-base reaches F1 72.1 (author-reported), while four off-the-shelf moderation tools reach at most F1 34.6. Other routes are crowd-labelled corpora, LLM-generated training data and model-in-the-loop data collection.ConceptTracking by detectionMulti-object tracking by detecting objects in every frame, then linking detections into identities. Covers a DeepSORT-style baseline, StrongSORT with offline linking, POI's pedestrian tracker, a 3D tracker on fused two-vehicle LiDAR, a sports tracker with visual similarity, and a clip-level transformer without linking. Trade-offs: appearance against motion, online against offline.ConceptTraining compute estimationHow to estimate the training compute of a model that reports none. Count network operations from the architecture, or hardware burden from processors, rate and time; a log-log conversion, fitted on ImageNet only, links them. Per-system estimates are expected within a factor of two. Operation counts are more precise; hardware burden is easier to find but depends on implementation efficiency.ConceptTraining data extractionAsks how much personal information a language model reveals from its training text through black-box queries. Covers observed sampling, a lazy estimator, prefix-and-suffix reconstruction and candidate-set inference, measured on fine-tuned GPT-2 with and without scrubbing and DP-SGD. Numbers are author-reported. DP-SGD at eps=8 cuts leakage sharply but not to zero.ConceptTrajectory optimizationChoosing a robot motion over time that minimises a cost while respecting joint, collision and torque limits. Covers the trajectory parameterisation, which limits are enforced inside the loop, and the trade-offs reported for a GPU-batched, dynamics-aware solver. One compiled card so far.ConceptTransfer learningStarting from weights learned on a source task and adapting them to a target task: by training a head on frozen features, fine-tuning all layers, adding adapters or columns, merging weights, or pre-training on unlabelled data. Fifty-seven cards cover these routes. Gains are author-reported; freezing and learning-rate details are often not stated.ConceptTrust-region methodMinimise a function, or search a black-box space safely, by trusting a local model only inside a region around the current guess. Two variants: a Gauss-Newton pose solver that sets its radius from the predicted-versus-actual decrease, and a safe Bayesian-optimisation box that grows or shrinks from success counts. Trade-offs: an extra subproblem solve per try, or many unstated constants.ConceptUnbalanced optimal transportOptimal transport in which mass may stay unmatched: each marginal is penalised (KL in the paper) instead of forced to equal the input mass, and an entropic term makes the plan solvable by Sinkhorn-type iterations. Gives a bounded response to outliers where balanced OT grows, and a debiased Sinkhorn divergence. One paper so far; domain-adaptation gains are author-reported.ConceptUncertainty estimationScoring how unsure a model is about an input, to rank unlabeled examples, flag unreliable answers or set an act-or-ask rule. Families: softmax margin and entropy (cheap, often overconfident), MC dropout with BALD, ensembles, GPs and SNGP, sampled-answer agreement, LLM self-grading, conformal sets, reject thresholds, per-box variance heads. Evidence is thin: eighteen cards, mostly author-reported.ConceptUnit-quaternion rotationA unit quaternion q represents a 3D rotation as q v q-bar. The papers use it in different roles: the unknown of a closed-form fit to vector pairs (smallest eigenvector of a 4x4 matrix), of a penalised non-linear fit with translation, or of a dual-quaternion pose fit solved by a trust region; or a network output renormalised after each step. Author-reported results do not compare across roles.ConceptUniversal approximationA large enough feedforward network can approximate a target to any accuracy. This page tracks what that costs in neurons and layers. Depth-based ReLU constructions give polylogarithmic size for smooth one-dimensional targets. RUMnet networks are dense in random utility choice models. These are existence results; only the RUMnet card gives a training bound.ConceptUnsupervised keypoint detectionLearns a few points or part regions per input without labels, trained through a downstream objective. Image animation: keypoints with local 2x2 Jacobians, thin-plate-spline warps over groups of five, or heatmap-moment regions for articulated bodies. Point clouds: the top-k points by feature norm, for registration. Author-reported numbers are not comparable across papers.ConceptUser simulationSimulated users stand in for real ones when testing or training dialogue and agent systems. A 2026 tutoring study finds that fine-tuned student simulators match real turns better than prompted ones, and a preference step adds a little more. A 2026 mobile-agent benchmark uses a hidden-profile LLM user to answer clarifying questions. Oracle-style prompts leak real answers.ConceptValue iterationFilling a table of the best expected return for each number of steps left and budget left, working back from the end, then reading a decision rule off the table. Exact cost grows as T times B. Large problems need an approximation: a small network for value differences, or backups only at sampled beliefs for partly observed problems.ConceptVariance-reduced gradientLowering the sample cost of a stochastic gradient by reusing an earlier estimate and correcting it with a small paired batch. Variants differ in how they reuse old samples: plain estimators need O(eps^-4) trajectories, reweighted recursions need a bounded-weight assumption, and defensive mixture sampling drops it at O(eps^-3) under score conditions.ConceptVariational autoencoderA variational autoencoder (VAE) encodes an input as a distribution over a latent and is trained with reconstruction plus a KL term to a prior. Variants add a condition, discrete codes (dVAE, VQ-VAE), adversarial or disentangling penalties, or a diffusion prior on the latent. Cards cover world models, zero-shot features, grasps, motion, video, depth, audio goals and design metrics.ConceptVariational inferenceVariational inference replaces an intractable posterior with a tractable family Q and maximises a lower bound on the marginal log-likelihood. Four cards cover variational EM for block models, the bound as a diffusion training objective, Bayes by Backprop and MC dropout, and amortised latent guidance in a recursive reasoner. Pick the variant by model structure, compute and sampling needs.ConceptVariational quantum circuitA small parameterised quantum circuit used as a trainable function. Bits encode the input, trained rotations and CNOT layers process it, and Pauli-Z expectations give the outputs. One card so far: a 4-qubit Q-network with 28 parameters, tested in simulation.ConceptVector quantizationVector quantization (VQ) snaps each vector to its nearest codebook entry, so a generator can predict discrete codes, a planner can use landmarks, or a stored set of vectors can shrink. Covers residual and multi-scale stacks, bit-wise codes, frozen tokenizers, action tokens and k-means codecs. Helps pick a variant by budget.ConceptVehicle platooningFollowers drive in a convoy at a fixed desired time gap behind a leader, tracking their predecessor's path. Vehicles join or leave the convoy through a platoon manager. The main variant choice is the join rule and the step size used in the following law. The only evidence so far is one simulation-only freeway study.ConceptVertical federated learningParties hold different features of the same users and train one model without pooling raw data. They exchange intermediate outputs and gradients after aligning users with private set intersection. Variants differ in the top model, whether the labelled party has features, and whether local models are trees. A federated RL survey uses "vertical" for another setting.ConceptVideo generationVideo from context frames, a GAN with continuous-time motion codes, conditioned video diffusion, a robot policy, joint audio and video diffusion, and image-model finetuning with correlated noise; plus a 17-metric text-to-video benchmark and a multiplayer world model. Covers FVD setups and the FVD results of six cards. All numbers are author-reported.ConceptVideo summarizationShortening a long video into still frames, a short clip or text that a person can review fast. The main choices are generic or question-driven selection, and the output form. One card so far, a 30-person study on 40-minute robot videos, with author-reported results.ConceptVirtual model control (VMC)Builds a robot controller from virtual springs, dampers and masses attached to robot points. Their forces map to joint torques through Jacobian transposes, so no inverse kinematics is needed. Six reaching and obstacle designs are given, tested on one humanoid arm. Results are author-reported and include no accuracy or optimality figures. One card so far.ConceptVision Transformer (ViT) backbonesA Vision Transformer cuts an image into patches, embeds each as a token and runs a Transformer encoder over them. Variants change the pretraining objective, the token mixer, the input or the role in a larger system. Frozen features often transfer to dense and specialised tasks. Trade-offs: results depend on pre-training, and attention cost grows with the square of the token count.ConceptVision-language pre-trainingVision-language pre-training ties images to text. Papers here train encoders with contrastive or align-then-fuse losses, bootstrap captions, train one model to read images and write text, bridge frozen encoders to LLMs, or use a frozen CLIP as a loss, prior or weight-space target. Main trade-off: data and compute cost against per-sample cost and licences.ConceptVision-language-action modelA policy that takes camera images and a language instruction and outputs robot actions, usually by building on a pretrained vision-language backbone. Covers action heads (tokens, regression, diffusion, flow matching), memory, cross-embodiment and generalist training, fine-tuning recipes, joint video-action models, failure detection, supervision and rollout evaluation.ConceptVisual groundingTying words to a place a system can act on. Shikra writes boxes as text (87.01 on RefCOCO val). TransVG regresses one box (81.02). GLaMM attaches a mask to each phrase. LM-Nav ranks map nodes. ScanQA ranks 3D proposals. GeoChat writes aerial boxes. Words2Contact picks a contact point. SpatialRGPT takes regions as input. RoboSpatial trains on free-space points. Figures are author-reported.ConceptVisual instruction tuningTraining a vision-language model on image or video instructions. Variants differ in the bridge and what trains: a linear projection (LLaVA), a Q-Former (InstructBLIP), an MLP with full LLM tuning (LLaVA-1.5), an abstractor (mPLUG-Owl), in-context examples (Otter), LoRA (PandaGPT), marks on images (ViP-LLaVA), coordinates as text (Shikra), tool calls (LLaVA-Plus) and driving decisions (DriveMLM).ConceptVisual object trackingFollowing one object (single-object tracking) or several with stable identities (multi-object tracking) through video. Covers eight paper cards: drone-benchmark SOT and MOT baselines, RGB-thermal adapters, a joint-ViT SOT tracker, sports, people-monitor, RNN, transformer and StrongSORT trackers. Trade-offs: accuracy against speed, motion against appearance, and the detector behind the tracker.ConceptVisual promptingTelling a vision model what matters in an input, usually without a new task head. Draw marks onto the image (ViP-LLaVA, VLMnav), tile worked examples into a grid (MAE-VQGAN), stack a solved pair (Painter), learn one frame added to every image for a frozen model (Bahng et al.), or learn prompt features or tokens inside the network (BAT, OVSeg).ConceptVisual question-answering evaluationScoring a vision-language model by asking it questions with known answers. Three uses: checking generated images against prompts (TIFA; Spearman 59.7 for mPLUG vs 33.2 for CLIPScore), benchmarking video models (VSI-Bench; best model 48.8 vs humans 79.2) and answering questions about 3D scans (ScanQA; EM@1 23.45 vs 22.31 for a 2D baseline). Costs one VQA call per question.ConceptVisual-inertial odometryVisual-inertial odometry estimates a platform's pose by fusing camera measurements with IMU readings. Kimera's VIO runs on a camera and IMU alone. Three cards use it inside lidar-inertial-visual systems: R2LIVE, LVI-SAM and R3LIVE. Accuracy is author-reported.ConceptVoice conversionChanging the emotion (or voice) of a recording while keeping the words. Emotional voice conversion needs parallel, multi-speaker emotional speech, which the ESD set provides. On ESD, CycleGAN-EVC beats VAWGAN-EVC on MCD and listening tests, at about ten times the training time. Author-reported, one speaker.ConceptWarp specializationWarp specialization gives each group of GPU threads one fixed job, such as loading data into shared memory or doing matrix multiplies and softmax, so that the jobs overlap. It needs Hopper-class hardware features. The only evidence so far is FlashAttention-3, where removing the role split or the matrix-softmax overlap each cost about 12-14% of throughput in one benchmark setting.ConceptWavelet transformSplits a signal into coarse scaling coefficients and detail (wavelet) coefficients at each scale, and rebuilds it from them. One card so far: a multiwavelet decomposition with fixed Legendre or Chebyshev filters, with small learned kernels per scale inside a PDE operator. Author-reported errors are lower than FNO's on most PDE tables here; the Chebyshev variant is weaker than Legendre.ConceptWeakly-supervised learningLearn from sparse, indirect or noisy supervision instead of clean full labels. Four cards: label-overlap loss for MR-to-ultrasound registration, mined web text pairs for embeddings, one label per slide for pathology, and a survey's taxonomy for 3D detection. Trade-off: cheap labels, but thin evidence, mostly author-reported.ConceptWeight decayA penalty on the size of the weights, added to the training loss. Covers pulling weights toward zero on a backbone or classifier, alone and with MaxNorm in long-tailed classification, decay on each feature net of an additive tabular model, and a squared anchor to source weights in source-free adaptation. Numbers are author-reported.ConceptWeight quantizationStoring weights in few bits on a grid. Per-channel 8-bit scales keep ImageNet CNNs near float accuracy. LLM.int8() keeps outlier features in 16-bit and stays within 0.2 perplexity of 32-bit. GPTQ corrects rounding error with calibration data, within 0.03 perplexity of FP16 for 4-bit OPT-175B. QLoRA's 4-bit NF4 keeps finetuned LLMs near 16-bit, but reports no speed gain.ConceptWord embeddingA word embedding maps each word to a dense vector learned from unlabelled text, so words in similar contexts get similar vectors. Variants differ in how the vector is learned (prediction, counts, character pieces or a language model), and so in whether they handle unseen words or context. Word-similarity test sets, scored with Spearman's rho, are the usual check and are small for most languages.ConceptWord embedding association testPermutation test of whether two target word sets lean toward two attribute word sets, giving a p-value. Gonen and Goldberg run it on two debiased embeddings and report all six p-values below 0.05. Wan et al. reuse the WEAT score on words from LLM-generated letters, with no p-values. Neither states every detail an implementer needs.ConceptWorld modelA learned model that predicts how a world changes after an action, so an agent can try options first. Variants: latent planners, hierarchical controllers, learned rewards, search over imagined futures, goal-conditioned models, driving forecasts, text simulators, models of other minds and typed shared state for multiplayer worlds. Physical consistency and long-rollout fidelity remain open.ConceptZero-shot learningAnswering for classes or tasks with no labelled examples. Routes: attribute or word-vector scores, generated pseudo-features, prompted LLMs and option scoring, CLIP-space 3D and adapters, text-to-region matching, event extraction, sketch, composed and video retrieval, robot policies, navigation agents, and attacks. Main trade-off: what supervision and compute you have.Conceptk-means clusteringPartition points into k groups by alternating nearest-centroid assignment and mean updates (Lloyd). Trails spectral clustering at default settings on synthetic benchmarks. Deletion variants drop one point at 2.2x to 584x lower amortized cost (loss ratio up to 1.44); a one-shot federated variant needs one upload per device. Author-reported.