Trusted Multi-view Learning for Long-tailed Classification

The introduction of the Trusted Multi-view Long-tailed Classification (TMLC) framework marks a significant advancement in addressing class imbalance in multi-view contexts, a challenge that has seen limited research. By focusing on long-tailed classification, TMLC employs a group consensus opinion aggregation mechanism inspired by Social Identity Theory, which directs decision-making towards the majority's preference. Additionally, it features an uncertainty-guided data generation module that enhances the quality of pseudo-data, effectively countering the negative effects of class imbalance. Extensive experiments conducted on long-tailed multi-view datasets have shown that TMLC achieves superior performance, underscoring its potential to transform classification tasks in artificial intelligence. The release of the code for TMLC further facilitates its adoption and encourages further research in this critical area.