R. Fu

Unlocking superior multimodal sentiment analysis, TSD reveals that disentangling features into common, pairwise, and private subspaces dramatically boosts performance.

By explicitly modeling a multi-level semantic hierarchy and carefully controlling information exchange between modalities, CLCR achieves state-of-the-art results in multimodal learning tasks ranging from emotion recognition to action recognition.