DecomVQANet: Decomposing visual question answering deep network via tensor decomposition and regression
作者:
Highlights:
• An optimal compression and acceleration solution for Visual Question Answering systems.
• The method reduces parameters about 80% while performance only drops 1%.
• The algorithm is a paradigm which can be widely used on most mobile devices.
摘要
•An optimal compression and acceleration solution for Visual Question Answering systems.•The method reduces parameters about 80% while performance only drops 1%.•The algorithm is a paradigm which can be widely used on most mobile devices.
论文关键词:Tensor decomposition,Tensor regression layer,Tensor contraction layer,Visual question answering
论文评审过程:Received 12 March 2020, Revised 21 May 2020, Accepted 3 July 2020, Available online 11 July 2020, Version of Record 1 November 2020.
论文官网地址:https://doi.org/10.1016/j.patcog.2020.107538