Reproduce “Dynamic Multimodal Information Bottleneck for Multimodality Classification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yingying Fang, Shuang Wu, Sheng Zhang, Chaoyan Huang, Tieyong Zeng, Xiaodan Xing, Simon Walsh, Guang Yang
Researchers developed the Dynamic Multimodal Information Bottleneck (DMIB) framework to improve multimodal classification robustness and performance, particularly in clinical contexts with noisy and redundant data. The framework consistently outperformed state-of-the-art methods across four medical datasets, demonstrating strong resilience to irrelevant or noisy modality inputs.
Key insights
- The information bottleneck principle can effectively distill a concise, informative feature representation by filtering noise and redundancy from multimodal inputs.
- A "sufficiency loss" is crucial for ensuring the information bottleneck retains task-relevant predictive power, preventing over-purification.
- Dynamic masking and individual modality supervision contribute to the framework's superior robustness and generalizability across diverse clinical data conditions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.