wikillm

Concept · 1 paper · updated 2026-10-19

Evidential deep learning

A 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.

uncertaintymulti-view-learningdempster-shaferevidential-learning

Markdown6 min read

AI-compiled overview, updated 2026-10-19, not yet reviewed by a human. Every claim cites its source; check it when a detail matters.

Overview

A standard classifier with a softmax output tends to be overconfident, and a model with several inputs usually weights each input the same way or with a fixed weight 2204.11423 §1. Evidential deep learning asks each network to report how much evidence it has for each class, instead of a probability alone. The evidence sets the parameters of a Dirichlet distribution over class probabilities 2204.11423 §3.1. From those parameters, the model derives a belief mass for each class and one uncertainty mass, and the masses always sum to one 2204.11423 §3.2.

The invariant that all variants share is this opinion structure: class beliefs plus one uncertainty mass, with total strength equal to the sum of the Dirichlet parameters. When an input has little evidence, its uncertainty mass is large, and its beliefs stay low. 2204.11423 §3.2

The paper extends this idea to several inputs ("views") of one sample. Each view gets its own opinion. The opinions are fused with a reduced form of the Dempster rule from evidence theory, so a view that is uncertain has less effect on the joint decision, and the joint decision has its own uncertainty value 2204.11423 §3.3.

Variants

Trusted multi-view classification (TMC): fuse per-view opinions

Each view has a network whose output is turned into evidence, and then into a per-view opinion 2204.11423 §3.1 2204.11423 §3.2. The views are fused pairwise into one joint opinion 2204.11423 §3.3. Training sums a loss term for each view and one for the fused opinion, and each term combines a fit part with a KL term that pulls the evidence toward zero for wrong classes 2204.11423 §3.1 2204.11423 §3.6. Card: Trusted Multi-View Classification with Dynamic Evidential Fusion.

On six vector datasets, TMC has the highest accuracy on Caltech101 (92.93) and is close to the best on Handwritten and CUB 2204.11423 tab. 1. Against the best single-view baseline on PIE and Scene15, the margin is larger: 91.99 against 84.36 on PIE, and 67.74 against 52.96 on Scene15 (the table's best single-view baseline on Scene15 is MCDO, not EDL) 2204.11423 tab. 1.

ETMC: TMC plus a pseudo-view

The extended model adds a pseudo-view, built from the concatenated features of the original views. It gets its own Dirichlet and enters the fusion in the same way 2204.11423 §3.5. The paper presents this as a way to model interaction between views at the representation level 2204.11423 §3.5.

ETMC has the top accuracy on five of the six vector datasets in the paper's table, with the largest gaps on PIE (93.75 against TMC's 91.99), Scene15 (71.61 against 67.74) and HMDB (69.43 against 65.26). On Caltech101, TMC is slightly higher 2204.11423 tab. 1. The extra parameters are about 0.02 million over TMC 2204.11423 tab. 5. On SUN RGB-D fusion, ETMC scores 61.3 against TMC's 60.7 2204.11423 tab. 3. Whether the pseudo-view itself is the cause is not isolated by an ablation in the results cited here.

Single-view evidential baseline

The paper trains a single-view evidential model as a baseline, labelled EDL 2204.11423 §4.1. Its accuracy is below both fused models on every vector dataset in the table, and its PIE and Scene15 scores are 84.36 and 46.41 2204.11423 tab. 1. The paper does not describe the baseline's construction in enough detail to reproduce it from this card alone; treat it as the authors' own implementation.

When to use which

SituationUseWhy
Several inputs describe one sample, and some may be noisy or corruptedTMCPer-input uncertainty lets an unreliable input count less in the fused decision 2204.11423 §1 2204.11423 tab. 1
Same as above, and you want the extra accuracy the pseudo-view givesETMCLargest margins over TMC on PIE, Scene15 and HMDB 2204.11423 tab. 1; about 0.02 million extra parameters 2204.11423 tab. 5
Little gain is needed on easy multi-view dataEither; check on your data firstThe paper's gains on Handwritten, CUB and Caltech101 are under one point over the best baseline 2204.11423 tab. 1
Tight compute budgetTMC or ETMC over a late-fusion networkThe fused models cost about the same as late fusion (3.64 GFLOPs, 22.37 million parameters), while the TRecg baseline needs 13.84 GFLOPs 2204.11423 tab. 5
Reject samples whose uncertainty is highThreshold on the fused uncertaintyOn SUN RGB-D, the paper's rule of dropping samples whose uncertainty exceeds the largest belief reduces the misclassification rate from 0.387 to 0.187 2204.11423 §4.2. This is the paper's own rule, not a standard calibration metric
Out-of-distribution inputs, simulated with Gaussian noise on test samplesFused uncertaintyNoisy test samples get higher uncertainty, and the effect grows with the noise level 2204.11423 §4.1
Calibrated probabilities are requiredCheck firstThe cited results report accuracy, AUROC and uncertainty behaviour, not a calibration error. Compare against calibration and ensembles before relying on the numbers
Only one input per sampleConsider Monte Carlo dropout or an ensembleThe fusion step that gives this method its advantage needs several inputs 2204.11423 §3.3. The single-view baseline is the authors' own run, not a validated replacement for a sampling method 2204.11423 tab. 1

All accuracy and uncertainty numbers above are author-reported.

Implementation sketch

The sketch follows the paper's equations for one training step 2204.11423 §3.6 with a fused opinion, written in Python style. Shapes are (B, K) for a batch of size B with K classes. Each step is annotated with its source.

Pseudocode
# Per-view evidence to Dirichlet parameters  [arxiv:2204.11423 §3.1]
def view_alpha(logits):
    evidence = softplus(logits)          # non-negative, (B, K)
    return evidence + 1                  # Dirichlet parameters, (B, K)

# Opinion from Dirichlet parameters  [arxiv:2204.11423 §3.2]
def to_opinion(alpha):
    strength = alpha.sum(-1, keepdim=True)   # S, (B, 1)
    belief = (alpha - 1) / strength          # (alpha_k - 1) / S
    uncertainty = K / strength               # K / S, (B, 1)
    return belief, uncertainty

# Reduced Dempster rule for two opinions  [arxiv:2204.11423 §3.3]
def combine(a, b):
    (ba, ua), (bb, ub) = a, b
    # Conflict: belief the two views put on different classes  [eq. 13]
    conflict = ba.sum(-1) * bb.sum(-1) - (ba * bb).sum(-1)
    # C = 1 divides by zero; the paper does not discuss this case
    scale = 1 - conflict
    belief = (ba * bb + ba * ub + bb * ua) / scale
    uncertainty = ua * ub / scale
    return belief, uncertainty

# Joint Dirichlet parameters from the fused opinion  [arxiv:2204.11423 §3.3]
def fused_alpha(belief, uncertainty):
    strength = K / uncertainty
    return belief * strength + 1
Pseudocode
# Loss for one opinion: fit term plus weighted KL to uniform  [§3.1]
def loss_term(onehot, alpha, lam):
    strength = alpha.sum(-1, keepdim=True)
    fit = (onehot * (digamma(strength) - digamma(alpha))).sum(-1)
    # Reset the true-class parameter to 1 before the KL  [eq. 7]
    alpha_t = onehot + (1 - onehot) * alpha
    # Closed-form KL to Dir(1, ..., 1)  [eq. 8]
    return fit + lam * kl_to_uniform(alpha_t)

# One training step: per-view and fused terms  [arxiv:2204.11423 §3.6]
def training_step(views, onehot, lam):
    alphas = [view_alpha(head(v)) for head, v in zip(heads, views)]
    fused = to_opinion(alphas[0])
    for alpha in alphas[1:]:
        fused = combine(fused, to_opinion(alpha))
    terms = [loss_term(onehot, a, lam) for a in alphas + [fused_alpha(*fused)]]
    return sum(terms).mean()   # the released code averages over the batch

Implementation notes:

  • KL weight. The paper says the weight increases gradually but gives no schedule 2204.11423 §3.1. The released code uses a linear ramp over a fixed number of epochs (paper card, Paper vs code). Early training depends on this choice.
  • Division by 1 - C. The fusion divides by 1 - C with no guard, so full conflict breaks the step. A small epsilon or a clamp is needed in practice.
  • Model selection. The released RGB-D scripts select checkpoints and the learning-rate schedule on the test loader, so those numbers may be optimistic. Use a separate validation split when comparing with other work (paper card, Pitfalls).
  • Scale of the loss. The paper writes sums over samples 2204.11423 §3.6. Averaging changes the balance against weight decay.
  • Vector demo. The paper describes a pseudo-view for vector data, but the released vector demo has none (paper card, Paper vs code).

Contradictions

  • Table 1 against table 2, same TMC runs. CUB accuracy is 91.00±0.42 in 2204.11423 tab. 1 and 90.83±3.23 in 2204.11423 tab. 2. PIE accuracy is 91.99±1.01 in tab. 1 and 91.91±0.11 in tab. 2. The standard deviations also differ for HMDB and Handwritten. Use tab. 1 with caution.
  • Percentage points. The text describes accuracy gains on PIE and Scene15 as percentages 2204.11423 §4.1. The values in the table imply percentage points.
  • Paper and released code. The paper gives weight decay of 1e-4 2204.11423 §4.1. The released scripts use 1e-5 (paper card, Paper vs code).

Papers

Related

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.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.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.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.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.

Source and license

Compiled from the papers listed on this page, all published under CC BY, CC BY-SA or CC0, with citations. See each paper's page for its authors and license.