Image Recognition Accuracy, Number of Parameters and Computational Complexity Using Channel Reduction by Dimensional Compression and Attention Function
Zhufeng Li, H. Yamauchi · 2023
Currently, due to the development of cloud services, artificial intelligence programs are being used in various places, and various machine learning models have been proposed. AI is also required to be realized on the edge side. However, on the edge side, the energy supply is severely limited, so an extremely energy-efficient AI is required. It is research. Currently, with the development of cloud services, artificial intelligence programs are used in various places, and various machine learning models have been proposed. There is a problem of time, and AI is required to be realized on the edge side as well. However, AI with extremely high energy efficiency is required due to the strong limitation of energy supply on the edge side. This paper is based on channel reduction by dimensional compression by GAP of SE blocks and channel attention function by weighting by sigmoid.