Reliable Disentanglement Multi-view Learning Against View Adversarial Attacks

TL;DR

A novel multi-view learning framework, namely Reliable Disentanglement Multi-view Learning (RDML), which proposes evidential disentanglement learning to decompose each view into clean and adversarial parts under the guidance of corresponding evidences, which is extracted by a pretrained evidence extractor.

Abstract

Trustworthy multi-view learning has attracted extensive attention because evidence learning can provide reliable uncertainty estimation to enhance the credibility of multi-view predictions. Existing trusted multi-view learning methods implicitly assume that multi-view data is secure. However, in safety-sensitive applications such as autonomous driving and security monitoring, multi-view data often faces threats from adversarial perturbations, thereby deceiving or disrupting multi-view models. This inevitably leads to the adversarial unreliability problem (AUP) in trusted multi-view learning. To overcome this tricky problem, we propose a novel multi-view learning framework, namely Reliable Disentanglement Multi-view Learning (RDML). Specifically, we first propose evidential disentanglement learning to decompose each view into clean and adversarial parts under the guidance of corresponding evidences, which is extracted by a pretrained evidence extractor. Then, we employ the feature recalibration module to mitigate the negative impact of adversarial perturbations and extract potential informative features from them. Finally, to further ignore the irreparable adversarial interferences, a view-level evidential attention mechanism is designed. Extensive experiments on multi-view classification tasks with adversarial attacks show that RDML outperforms the state-of-the-art methods by a relatively large margin. Our code is available at https://github.com/Willy1005/2025-IJCAI-RDML.

Figures (5)

• Figure 1: We conduct experiments on the PIE dataset in clean and adversarial settings, and show a toy example of the AUP. Note that only one view is randomly attacked via PGD. (a) presents the estimated uncertainties of the ECML method and our RDML. (b) shows the classification accuracy of these methods.
• Figure 2: The framework of RDML. (i) Evidential disentanglement learning uses the pretrained evidence extractor $E_{pt}(\cdot)$ to conduct a robustness analysis of features under random view attacks, and generate a robustness mask to decouple clean and adversarial features. (ii) The feature recalibration module will rectify the adversarial features. For features that are difficult to repair, RDML will generate evidential attention with the guidance of $E_{pt}(\cdot)$ to further mitigate the interference of adversarial features. (iii) RDML introduces the Dempster Rule-based Fusion for opinion aggregation.
• Figure 3: Density of estimated uncertainty on MSRC with different numbers of attacked views.
• Figure 4: Classification accuracy (%) on four datasets with different $\gamma_{1}$ and $\gamma_{2}$ (one random view is attacked).
• Figure 5: Classification accuracy (%) on MSRC with different numbers of attacked views.