MMBypass: Towards efficient multi-modal AI computing with adaptive bypass network

Yifei Pu, Xinfeng Xia, Xiaofeng Hou, Chi Wang, Cheng Xu, Jiacheng Liu, Jing Wang, Minyi Guo, Jingling Yuan, Chao Li

Abstract: Highlights•Propose a new neural network architecture for efficient multi-modal ai computing.•Design two lightweight modules for low execution latency while keeping accuracy.•Evaluate on real-world multi-modal tasks and reduce up to 44.5% execution latency.•Apply parallel computing to multi-modal models and compare the performance.