Navigating Conflicting Views: Harnessing Trust for Learning

Jueqing Lu, Lan Du, Wray Buntine, Myong Chol Jung, Joanna Dipnall, Belinda GabbeSubmission to ICLR 2025 · Rejected

Abstract

Resolving conflicts is essential to make the decisions of multi-view classification more reliable. Much research has been conducted on learning consistent and informative representations among different views, often assuming that all views are equally important and perfectly aligned. However, real-world multi-view data may not always conform to these assumptions, as some views may express distinct information. To address this issue, we develop a computational trust-based discounting method to enhance the existing Evidential Multi-view framework in scenarios where conflicts between different views may arise. Its belief fusion process considers the reliability of predictions made by individual views via an instance-wise probability-sensitive trust discounting mechanism. We evaluate our method on six real-world datasets, using Top-1 Accuracy, Fleiss’ Kappa, and a new metric called Multi-View Agreement with Ground Truth that takes into consideration the ground truth labels, to measure the reliability of the prediction. We also evaluate whether uncertainty measures can effectively indicate prediction correctness by calculating the AUROC. The experimental results show that computational trust can effectively resolve conflicts, paving the way for more reliable multi-view classification models in real-world applications.
BibTeX
@inproceedings{lu2025navigating,
  author = {Jueqing Lu and Lan Du and Wray Buntine and Myong Chol Jung and Joanna Dipnall and Belinda Gabbe},
  title = {{Navigating Conflicting Views: Harnessing Trust for Learning}},
  booktitle = {International Conference on Machine Learning},
  year = {2025},
  url = {https://openreview.net/forum?id=uJ3JqtBYWk},
}
Figure