Asynchronous Multimodal Video Sequence Fusion via Learning Modality-Exclusive and -Agnostic Representations | Synapse