Before Fusion, Ask What to Keep: Contextual Calibration of Multimodal Signals

Jiyuan Liu, Liangwei Nathan Zheng, Wei Emma Zhang +2

Multimodal systems often benefit from combining information across language, sound, and visual streams, but this benefit is not guaranteed. A modality that is useful for one input…