MuteBench: Modality Unavailability Tolerance Evaluation for Incomplete Multimodal Fusion
MuteBench is a benchmark designed to test how well clinical AI systems using multiple data sources (modalities) handle sensor failures. Researchers tested six fusion architectures against two failure types: missing an entire channel or losing time segments within a channel. Findings show that the architecture family matters most, and specific imputation techniques can help improve performance when data is missing.