[Paper Review] Trusted Multi-View Classification

Zongbo Han, Changqing Zhang+2 more|arXiv (Cornell University)|Feb 3, 2021

Adversarial Robustness in Machine LearningComputer Science47 references49 citations

TL;DR

The paper presents a trusted multi-view classification method that fuses view-specific evidences via Dirichlet distributions and Dempster-Shafer theory to produce reliable uncertainty estimates and robust predictions, including out-of-distribution detection.

ABSTRACT

Multi-view classification (MVC) generally focuses on improving classification accuracy by using information from different views, typically integrating them into a unified comprehensive representation for downstream tasks. However, it is also crucial to dynamically assess the quality of a view for different samples in order to provide reliable uncertainty estimations, which indicate whether predictions can be trusted. To this end, we propose a novel multi-view classification method, termed trusted multi-view classification, which provides a new paradigm for multi-view learning by dynamically integrating different views at an evidence level. The algorithm jointly utilizes multiple views to promote both classification reliability and robustness by integrating evidence from each view. To achieve this, the Dirichlet distribution is used to model the distribution of the class probabilities, parameterized with evidence from different views and integrated with the Dempster-Shafer theory. The unified learning framework induces accurate uncertainty and accordingly endows the model with both reliability and robustness for out-of-distribution samples. Extensive experimental results validate the effectiveness of the proposed model in accuracy, reliability and robustness.

Motivation & Objective

• Motivate the need for uncertainty-aware multi-view learning in safety-critical scenarios.
• Propose a unified framework that integrates multi-view evidence at the level of beliefs rather than features or outputs.
• Enable accurate per-view and joint uncertainty estimation to improve reliability and robustness, including OOD detection.
• Show that the approach achieves superior accuracy, reliability, and robustness across multiple datasets.

Proposed method

• Model each view with evidence that parameterizes a Dirichlet distribution over class probabilities.
• Use subjective logic to relate view evidence to per-view belief masses and an uncertainty mass.
• Combine per-view beliefs using Dempster–Shafer theory to obtain a joint belief and Dirichlet parameters.
• Convert the joint belief into Dirichlet parameters (alpha) to derive final class probabilities and uncertainty.
• Train networks to output nonnegative evidence vectors via a loss that combines an adjusted cross-entropy (ACE) and a KL divergence term to suppress incorrect evidence.
• Optimize in a multi-task fashion across all views with a joint loss that aggregates per-view and joint objectives.

Experimental results

Research questions

• RQ1 How can multi-view information be fused at the evidence level to produce reliable predictions and uncertainty estimates?
• RQ2 Can Dirichlet-based evidence modeling combined with Dempster-Shafer fusion improve robustness to noisy or out-of-distribution views?
• RQ3 Does the proposed framework provide accurate per-view and joint uncertainty useful for trusted decision making?

Key findings

• Out-of-sample accuracy and AUROC exceed those of single-view uncertainty methods across six datasets.
• The method maintains strong performance under noisy multi-view conditions due to view-specific uncertainty awareness.
• Joint uncertainty and class probabilities can be inferred from fused Dirichlet evidence, enabling reliable trust signals for predictions.
• Uncertainty tends to be higher for out-of-distribution samples, supporting effective OOD detection.
• The approach shows notable improvements on hard-scene and action recognition datasets (e.g., Scene15, HMDB).