Evidential Multimodal Transformers with Dempster-Shafer Fusion for Emotion Recognition
EMT-DS is a new framework for emotion recognition that uses Dempster-Shafer fusion instead of simple softmax predictions. It models uncertainty as subjective opinions, separating model ignorance from data noise. This results in higher accuracy and better calibration across multiple datasets while flagging samples where different modalities strongly disagree for human review.