Research papers, explained for engineers and their coding agents
Each paper becomes a page you can build from: the idea in plain words, the method step by step, every hyperparameter, the results, and where the official code differs from the paper. Every claim links to its source.
1653 papers · 718 concepts
Paper of the day, 2026-11-01
ManipLLM: Embodied Multimodal Large Language Model for Object-Centric Robotic Manipulation
Concepts
The ideas that the most papers build on. Start here to compare the approaches to a problem.
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.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.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.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.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.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.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.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.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.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.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.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.
Latest papers
Newest first.
Paper · 2026-10-20Reinforcement learning for hardware acceleration and system-level optimization on FPGA and integrated-circuit platforms: A surveySurvey of more than 100 studies running reinforcement learning on FPGAs and integrated circuits. It sorts them by RL family (MDP, Q-learning, SARSA, DQN, DDPG, actor-critic, Monte Carlo, PPO, multi-agent), by RL role (workload or optimizer) and by application. No new experiments; the reported best results, such as 1400x for one Q-learning design, are author-reported.Paper · 2026-10-14Object Pose from 2-D to 3-D Point and Line CorrespondencesEstimates the rotation and translation of a camera relative to a 3-D object from point and/or line correspondences. Each correspondence gives quadratic constraints on a dual-number quaternion; a penalised sum of squares is minimised with a trust-region method. With 100 points and 1 px noise, author-reported translation error is 0.0002 cm against 11.48 cm for the linear Faugeras-Toscani method.Paper · 2026-10-13Calibrated Disambiguation for Partial Multi-label LearningPML-CD trains a multi-label image classifier on over-annotated labels by weighting each candidate positive with a calibrator. The calibrator maps a 20-bin histogram of confidences on candidate positives to per-bin reliability, and an extra prototype-cosine loss is added. Author-reported mAP beats nine baselines in 15 of 16 settings on VOC, COCO, NUS-WIDE and CUB at 20-80% noise.Paper · 2026-10-12Boosting the Learning for Ranking PatternsLearns a user's ranking of mined patterns as weights over interestingness measures. Pairwise measure gaps against user rankings feed an AHP eigenvector solve. An active mode asks pairwise questions chosen by a sensitivity heuristic. On association rules (author-reported) it is far faster than an SVM ranker and more accurate in active mode, but less accurate in passive mode.Paper · 2026-10-12Learning to Play: Reinforcement Learning and GamesTextbook on reinforcement learning in games. Covers MDPs, heuristic search, Monte Carlo tree search with UCT, deep RL and self-play, using AlphaGo, AlphaGo Zero and AlphaZero as case studies. Gives the MCTS loop and the reported AlphaGo settings. Author-reported: AlphaGo beat Fan Hui, Lee Sedol and Ke Jie; AlphaZero learned Go, Chess and Shogi from self-play.Paper · 2026-10-11Data Portability Revisited: Toward the Human-Centric, AI-Driven Data Ecosystems of TomorrowA legal article arguing that EU data-portability rules (GDPR Art. 20, Data Act, DMA, EHDS, AI Act) mainly serve service providers. It compares three architectures (consent and opt-out, enterprise-centric pods and GAIA-X, user-held private-by-default data) and urges treating portability as a fundamental right. It has no experiments or code; its only numbers are regulatory parameters.Paper · 2026-10-05Virtual model control for compliant reaching under uncertaintiesBuilds reaching controllers for a force-controlled arm from virtual springs and dampers, mapped to joint torques through Jacobians, with no inverse kinematics. Six designs cover simple, moving-reference and force-driven reaching, plus obstacle avoidance. A human blocking the hand on a Sciurus17 humanoid drew up to 4.53 N, versus 35.77 N for an IK-based PID baseline (author-reported).Paper · 2026-10-02Sample complexity of variance-reduced policy gradient: weaker assumptions and lower boundsProposes DEF-PG, a variance-reduced policy gradient using defensive importance sampling, which keeps importance weights bounded without a uniform variance assumption. Under bounded rewards and regular policy scores it reaches O(eps^-3) first-order sample complexity. It also proves black-box lower bounds of Omega(eps^-4) for one-policy and Omega(eps^-3) for coupled feedback.Paper · 2026-10-01The Impact of Processing Parameters on High-Accuracy Measurements in UAV PhotogrammetryFull-factorial study of 768 Bundle Block Adjustment (BBA) processing variants in Agisoft Metashape on ten UAV surveys (220 ha). Number of GCPs, extra camera-calibration corrections and PPK georeferencing matter most. Best variant: 16 mm 3D RMSE at check points, worst 303 mm. Tuned settings halve the systematic vertical error (14 to 7 mm). Author-reported.Paper · 2026-09-30Recognition of Urbanized Areas in UAV-Derived Very-High-Resolution Visible-Light ImageryCompares 16 RGB and CIELab vegetation indices, thresholded on one UAV dataset, with linear, MLP and CNN pixel classifiers for urbanized vs non-urbanized areas in UAV RGB orthomosaics. Tested on a second site with another drone and season. Best network: MCC about 0.82, accuracy about 0.96. Best index, Excess Blue: MCC about 0.61, accuracy about 0.87. Author-reported.Paper · 2026-09-29Learning from synthetic photorealistic raindrop for single image raindrop removalRenders a physics-based synthetic dataset of adherent raindrops on Cityscapes images, with pixel-level masks, and trains a three-stage network: raindrop mask detection, region reconstruction, and refinement. Author-reported PSNR 41.29 dB and SSIM 0.9921 on the synthetic test set, above the compared baselines.Paper · 2026-09-28Word Similarity Datasets for Indian Languages: Annotation and Baseline SystemsBuilds manually annotated word-similarity test sets for six Indian languages (Urdu, Telugu, Marathi, Punjabi, Tamil, Gujarati) by translating and re-annotating RG-65 and WordSim-353, then scores Skip-gram, CBOW and FastText embeddings with Spearman's rho on three of them. FastText with an OOV fallback reaches 45.5 (Urdu), 38.7 (Marathi) and 46.0 (Telugu). Author-reported.
For coding agents
Every page is also plain markdown: add .md to any page URL. Agents can start at llms.txt, query the search API, or connect the MCP server at https://wikillm.com/mcp.