Robust Fuzzy Multi-View Learning Framework Addresses View Conflict
A novel approach known as Robust Fuzzy Multi-View Learning (R-FUML) has been introduced to address view conflicts in multi-view classification. Current trusted multi-view classification (TMVC) techniques rely on the unrealistic assumption of perfect alignment between views, which does not hold true in practical situations. R-FUML leverages Fuzzy Set Theory to represent network outputs as fuzzy memberships, allowing for the assessment of category credibility. Additionally, it employs an entropy-based strategy to alleviate problems such as underestimated uncertainty, erroneous decisions, and overfitting due to view conflicts. This research is documented in the arXiv preprint 2605.24475.
Key facts
- R-FUML is a novel framework for multi-view classification under view conflict.
- Existing TMVC methods assume strict alignment across views.
- R-FUML uses Fuzzy Set Theory to model outputs as fuzzy memberships.
- It addresses underestimated uncertainty, misleading decisions, and overfitting.
- The research is published on arXiv with ID 2605.24475.
- The paper extends TMVC to handle view conflict during training and inference.
- R-FUML employs an entropy-based method to mitigate view conflict.
- The framework aims to deliver reliable fusion for accurate predictions.
Entities
Institutions
- arXiv