Fairness-Aware Multi-view Evidential Learning with Adaptive Prior
Haishun Chen, Cai Xu, Jinlong Yu, Yilin Zhang, Ziyu Guan, Wei Zhao, Fangyuan Zhao, Xin Yang
TL;DR
This paper introduces FAML, a fairness-aware multi-view evidential learning method that uses adaptive priors and fairness constraints to improve evidence balance, prediction accuracy, and uncertainty reliability.
Contribution
FAML is the first to incorporate adaptive priors and explicit fairness constraints into multi-view evidential learning for unbiased evidence allocation.
Findings
FAML achieves more balanced evidence distribution across classes.
FAML improves prediction accuracy and uncertainty reliability.
FAML outperforms state-of-the-art methods on real-world datasets.
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…
Peer Reviews
Decision·ICLR 2026 Poster
1. The paper has a clear motivation and effectively solves the problems of biased evidence multi-view learning. 2. The paper offers a clear and well-grounded theoretical analysis that connects the adaptive prior design to margin theory, helping explain why the proposed approach could improve model's generalization. 3. The comparison experiments are comprehensive, including six representative multi-view datasets.
1. This work proposes an EDL-based multi-view classification method. However, the literature review for existing EDL-based multi-view methods is insufficient. The authors should provide a more comprehensive discussion of related work in this specific domain, such as, but not limited to, [1, 2]. 2. The text in Figure 1 is too small, and there is no explanation of what the points, lines, and colors in the figure represent or why it is imbalanced. The blue in the legend of Figure 1 is different fr
1. This paper is well organized, and the proposed methodology is enlightening. 2. The motivation behind the paper is clear, and the theoretical analysis is complete. 3. The proposed method offers novel insights, particularly in using training trajectories to adjust class priors, thereby mitigating view-specific bias throughout the multi-view fusion process 4. The proposed method shows a clear performance improvement in a series of experiments.
1. In this paper, the notion of fairness seems to focus on balancing the evidence allocation across different classes, rather than addressing fairness in terms of sensitive attributes like race or gender in a broader sense. 2. Is this approach intended as a general framework? Can other trusted multi-view fusion methods also adopt similar strategies to improve model performance even on balanced datasets? 3. Some implementation details seem to be missing. For instance: How does the hyper-paramet
To address this, the paper proposes a Fairness-Aware Multi-view Evidential Learning (FAML) framework. The method has three core components: An Adaptive Prior based on training trajectories, which dynamically adjusts the Dirichlet prior to provide more support to classes that are under-represented or poorly performing. An explicit Fairness Constraint, which penalizes high variance in evidence allocation across different classes, encouraging a more balanced evidence distribution. An Opinion Ali
While this is an excellent paper, the following minor revisions could further strengthen its clarity and impact: Strengthen the "Related Work" on Fairness and Subpopulation Shift: The core problem FAML solves—evidence bias due to class imbalance—is deeply connected to the broader fields of AI fairness and subpopulation shift. To better position the paper's contribution, the "Related Work" section should be expanded to include and discuss key works from this area. For instance, the authors shoul
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsDomain Adaptation and Few-Shot Learning · Stochastic Gradient Optimization Techniques · Advanced Bandit Algorithms Research