Evidential dissonance measure in robust multi-view classification to resist adversarial attack

Xiaodong Yue, Zhicheng Dong, Yufei Chen, Shaorong Xie

Abstract: Highlights•Propose a dissonance measure to evaluate data views subjected to adversarial attacks.•Propose a dissonance-aware belief integration strategy for multi-view fusion.•Analyze the properties of the dissonance-aware belief integration strategy.•Implement a robust multi-view classification method with belief integration strategy.•Sufficient experiments verify the model robustness against adversarial attacks.