Fairness-Aware Multi-view Evidential Learning with Adaptive Prior

TL;DR

The paper tackles biased evidence allocation in multi-view evidential learning, where data-rich classes tend to dominate evidence and distort uncertainty. It introduces FAML, which uses a training-trajectory-based adaptive Dirichlet prior, a class-wise fairness constraint based on evidence variance, and an opinion-alignment fusion mechanism to harmonize cross-view evidence. Theoretical analysis via margin theory shows the adaptive prior improves minority-class margins and tightens generalization bounds, while extensive experiments on five real-world datasets demonstrate improved accuracy and more reliable uncertainty estimates with balanced evidence across classes. This approach yields practical benefits for high-stakes applications by delivering fairer uncertainty and robust predictions across diverse data regimes.

Abstract

Multi-view evidential learning aims to integrate information from multiple views to improve prediction performance and provide trustworthy uncertainty esitimation. Most previous methods assume that view-specific evidence learning is naturally reliable. However, in practice, the evidence learning process tends to be biased. Through empirical analysis on real-world data, we reveal that samples tend to be assigned more evidence to support data-rich classes, thereby leading to unreliable uncertainty estimation in predictions. This motivates us to delve into a new Biased Evidential Multi-view Learning (BEML) problem. To this end, we propose Fairness-Aware Multi-view Evidential Learning (FAML). FAML first introduces an adaptive prior based on training trajectory, which acts as a regularization strategy to flexibly calibrate the biased evidence learning process. Furthermore, we explicitly incorporate a fairness constraint based on class-wise evidence variance to promote balanced evidence allocation. In the multi-view fusion stage, we propose an opinion alignment mechanism to mitigate view-specific bias across views, thereby encouraging the integration of consistent and mutually supportive evidence.Theoretical analysis shows that FAML enhances fairness in the evidence learning process. Extensive experiments on five real-world multi-view datasets demonstrate that FAML achieves more balanced evidence allocation and improves both prediction performance and the reliability of uncertainty estimation compared to state-of-the-art methods.

Figures (5)

• Figure 1: Visualization of biased evidence learning and uncertainty estimation on BRCA dataset. Left: Evidence space with training samples (dots) and categorical decision boundaries (in shade) for Her2, Normal, and Basal classes in two views. Right: Corresponding uncertainty estimation, where darker regions indicate the prediction is more uncertain.
• Figure 2: Illustration of FAML.We first employ view-specific evidential neural networks to construct opinions from each view. To mitigate evidential bias, a training-trajectory-based adaptive prior is introduced during evidence learning, ensuring fairer evidence allocation. An opinion alignment mechanism is then applied to promote consistency among different views. Finally, all view-specific opinions are integrated to reach a reliable and trustworthy decision.
• Figure 3: Uncertainty Estimation on Animal dataset.
• Figure 4: The average evidence strength of each category on Handwritten dataset.
• Figure 5: Accuracy(%) when adjusting $\gamma$ on datasets.

Theorems & Definitions (3)

• Definition 1: Evidence Margin
• Lemma 2: Margin Generalization Bound
• Theorem 3