---
title: Conflict-aware Evidential Deep Learning
url: https://www.emergentmind.com/topics/conflict-aware-evidential-deep-learning-c-edl
type: topic
---

# Conflict-aware Evidential Deep Learning

Conflict-aware Evidential Deep Learning (C-EDL) encompasses a family of methods designed to improve the robustness and uncertainty quantification of Evidential Deep Learning (EDL) models, especially under data regimes characterized by conflicting, incomplete, adversarial, or out-of-distribution (OOD) evidence. These approaches explicitly measure representational disagreement—either among input transformations or multi-view observations—and calibrate uncertainty scores and predictions accordingly. C-EDL instantiations include lightweight, post-hoc adjustments for pre-trained EDL classifiers as well as integrated, architecture-level conflict resolution within multi-view settings. They achieve state-of-the-art performance in OOD and adversarial detection, and in robust multi-view classification under missingness and view corruption [2409.06270] [2506.05937].

## 1. Evidential Deep Learning: Foundations and Limitations

EDL provides a non-Bayesian framework for uncertainty estimation by modeling class predictions as parameters of a Dirichlet distribution. Given input $x \in \mathbb{R}^d$ and $K$ classes:
- The EDL network predicts non-negative evidence $e_k(x) \geq 0$. Dirichlet parameters $\alpha_k = e_k(x) + 1$.
- The total evidence $S(x) = \sum_{k=1}^K \alpha_k$.
- The belief mass $b_k(x) = \frac{e_k(x)}{S(x)}$ and the remaining uncertainty $u(x) = \tfrac{K}{S(x)}$.
- The expected predictive probability $E[p_k] = \alpha_k / S(x)$.

Training minimizes an “evidential loss” that combines a data fidelity term and a KL-divergence regularizer toward the Dirichlet uniform prior, penalizing overconfident mispredictions.

However, standard EDL is vulnerable to overconfident misclassification in OOD and adversarial settings. A single forward pass yields high evidence even for invalid or corrupted inputs, and traditional Dempster–Shafer (DS) fusion amplifies this problem in multi-view or transformation-ensemble scenarios due to its sensitivity to conflict [2506.05937] [2409.06270].

## 2. Conflict in Evidential Fusion: Formalization and Impact

In DS theory, each evidence source provides a basic belief assignment (BBA) $m: 2^\Theta \rightarrow [0,1]$ over the frame of discernment $\Theta$. The DS combination rule fuses two BBAs $m^1$ and $m^2$ as:
- The conflict mass $K = \sum_{A \cap B = \emptyset} m^1(A) \cdot m^2(B)$.
- The fused mass for $C \neq \emptyset$, $m(C) = \frac{1}{1-K} \sum_{A \cap B = C} m^1(A) m^2(B)$.

When sources are highly contradictory ($K \to 1$), the denominator $1-K$ approaches zero, causing numerical instability, unpredictable magnification of uncertainty, and degraded uncertainty quality. In incomplete multi-view classification, imputation errors can induce frequent moderate-to-severe conflicts during evidence fusion, undermining confidence calibration and model reliability [2409.06270].

This motivates conflict-aware mechanisms that explicitly measure and respond to evidence disagreement.

## 3. Conflict-aware Dempster–Shafer Combination Rule (DSCR) in Multi-View Learning

The Alternating Progressive Learning Network (APLN) and its DSCR represent a principled integration of conflict-aware EDL in multi-view, incomplete-data regimes [2409.06270]. The DSCR operates as follows:
- For singleton opinions $\omega^A = (b^A, u^A, a)$ and $\omega^B = (b^B, u^B, a)$, conflict is $K = \sum_{i \neq j} b^A_i \cdot b^B_j$.
- The unnormalized fused belief and uncertainty are:
  - $\tilde{b}_k = \frac{b^A_k u^B + b^B_k u^A}{u^A + u^B}$
  - $\tilde{u} = \frac{2u^A u^B}{u^A + u^B}$
- Both $\tilde{b}_k$ and $\tilde{u}$ are down-weighed by $(1-K)$: $\bar{b}_k = (1-K)\tilde{b}_k$, $\bar{u} = (1-K)\tilde{u}$
- Final normalization: $Z = \sum_k \bar{b}_k + \bar{u}$, $b^{A \oplus_{CA} B}_k = \bar{b}_k/Z$, $u^{A \oplus_{CA} B} = \bar{u}/Z$

This formulation ensures that strong conflict (large $K$) shrinks fused belief evidence and transfers mass to uncertainty, preventing instability and yielding a more reliable combined opinion under incompleteness and conflict.

Within APLN, multi-view data proceeds through three learning phases: coarse imputation and latent alignment (UMAE-F), progressive evidence learning (UMAE-V), and joint end-to-end optimization (UMAE-J), always fusing evidence via DSCR to stabilize both training and inference.

## 4. Post-hoc Conflict-Aware EDL (C-EDL) for OOD and Adversarial Detection

A distinct instantiation of C-EDL provides a post-hoc, lightweight approach for uncertainty calibration in standard EDL classifiers without retraining [2506.05937]. The method operates as follows:
- For each test sample $x$, generate $T$ task-preserving, metamorphic transformations $\{\tau_t(x)\}$.
- For each transformed input, obtain Dirichlet parameters $\alpha^{(t)}$ from the pre-trained EDL network.
- Compute conflict over the evidence set $\mathcal{A} = \{\alpha^{(1)}, ..., \alpha^{(T)}\}$ using:
  - Intra-class variability: $C_n = \frac{1}{K} \sum_{k=1}^K \frac{\mathrm{std}(\{\alpha^{(t)}_k\})}{\mathrm{mean}(\{\alpha^{(t)}_k\}) + \epsilon}$
  - Inter-class contradiction: $C_i = \frac{1}{T} \sum_t [1 - \exp(-\beta \cdot \sum_{k<j} (...))]$ (see source for full detail)
  - Total conflict $C = C_i + C_n - C_i C_n - \lambda (C_i - C_n)^2$, with $\lambda \in [0, 1/2]$
- Average and decay evidence via $\bar{\alpha}_k = \frac{1}{T}\sum_t \alpha^{(t)}_k$, $\tilde \alpha_k = \bar{\alpha}_k \exp(-\delta C)$.
- Compute final uncertainty $\tilde u = K / \sum_k \tilde \alpha_k$ and use it for abstention or OOD/adversarial flagging: predict if $\tilde u \leq \tau$; otherwise abstain.

This framework uses conflict as a trigger to reduce posterior confidence, selectively elevating uncertainty where transformations strongly disagree, and suppressing overconfident errors on anomalous inputs.

## 5. Optimization and Loss Formulations

In the multi-view APLN setting with DSCR [2409.06270], the objective comprises:
- EDL evidence loss: $\mathcal{L}_{ACE}(\alpha_n) = \sum_{j=1}^K y_{nj} [\psi(S_n) - \psi(\alpha_{nj})]$ (with digamma $\psi$)
- KL divergence to a uniform Dirichlet: $\mathcal{L}_{KL}$
- Conflict consistency loss: for every view pair $(A, B)$, compute a divergence $D_{JS}(q^A \| q^B)$, and optimize the mean conflict degree $c(\omega^A, \omega^B) = 1 - D_{JS}(q^A \| q^B)$, then $\mathcal{L}_{con} = \frac{1}{V-1}\sum_{A,B \neq A} c(\omega^A, \omega^B)$
- ELBO regularization for latent imputation using a VAE

Sampling for missing views in APLN is stochastic: the VAE samples $L$ latent codes for each missing view, evidence is averaged before forming Dirichlet parameters, thus propagating uncertainty from missingness explicitly into the fused predictive distribution.

For post-hoc C-EDL [2506.05937], only inference phase computation is required, and no training loss modification is imposed.

## 6. Empirical Validation and Comparative Performance

C-EDL methods achieve consistent state-of-the-art performance across both incomplete multi-view settings and OOD/adversarial detection tasks.

In incomplete-view multi-view classification [2409.06270]:
- Datasets: YaleB, Handwritten, ROSMAP, BRCA, Scene15, NUS-Wide.
- When missingness rate increases ($\eta$ up to 0.5), APLN+DSCR achieves highest accuracy (e.g., Handwritten at $\eta=0.4$: 97.05% vs UIMC’s 97.00%; ROSMAP at $\eta=0.5$: 72.97% vs 71.43%).
- On conflict test splits (e.g., 40% of samples with cross-class view swaps), APLN+DSCR maintains accuracy within 1–2% of non-conflict performance, while standard DS fusion suffers accuracy drops up to 5%.
- Uncertainty metrics improve (average $u$ decreases with more coherent evidence), and accuracy variance is reduced.

In OOD and adversarial detection [2506.05937]:
- Across MNIST and CIFAR-10 tasks, C-EDL retains >94% ID coverage, reduces OOD coverage by up to 55%, and adversarial coverage by up to 90% compared to standard EDL.
- For example, under severe attack (MNIST$\rightarrow$FashionMNIST, L2-PGD, $\epsilon=1.0$): EDL retains 52.21% of adversarial samples, C-EDL only 15.51%; ID coverage remains high (EDL 96.61%, C-EDL 94.18%).
- Runtime overhead is minimal ($\sim4\times$ EDL), as only T ($\sim5$) forward passes and lightweight evidence statistics are required.

| Method         | ID Coverage | OOD Coverage | Adv Coverage (L2-PGD, $\epsilon=1$) |
|----------------|-------------|--------------|-------------------------------------|
| Standard EDL   | 96.61%      | 2.52%        | 52.21%                              |
| C-EDL          | 94.18%      | 1.77%        | 15.51%                              |

These results demonstrate that C-EDL mechanisms, whether integrated (DSCR) or post-hoc, robustly prevent overconfident false predictions in high-conflict, high-uncertainty, and adversarial contexts without sacrificing in-distribution performance.

## 7. Significance and Theoretical Implications

Conflict-aware Evidential Deep Learning establishes a general methodology for enhancing epistemic uncertainty quantification in neural models:
- By explicitly quantifying and attenuating conflict, it stabilizes evidence aggregation in both multi-view and transformation-based ensembles.
- It is agnostic to architecture and can be used either as a training-integrated module (as in DSCR/APLN) or inference-only post-processing (as in C-EDL).
- The theoretical formulations remain consistent with Dempster–Shafer subjective logic, and provide analytic conflict metrics with monotonicity guarantees.
- The negligible computational overhead and the empirical state-of-the-art improvement in both robustness and uncertainty calibration distinguish C-EDL as a general-purpose uncertainty amplification strategy.

A plausible implication is that conflict quantification—via statistical divergence measures or DS-style combiners—could become standard for uncertainty adjustment in other evidential and Bayesian deep learning domains, especially as deployment in real-world safety-critical applications increases [2409.06270] [2506.05937].

Source: https://www.emergentmind.com/topics/conflict-aware-evidential-deep-learning-c-edl