Microarchitecture searching method and device for deep convolutional neural network

• Main Authors: HU QINGHAO, SHAO SHUAI, LIU ZHAO, WANG SIYUAN, HU XURAN, CHEN XIANDA, XU MINGKAI, GAO FEI, YANG NING, GAO XIANG, ZHANG YUWEI, LI CONG, LIU YINAN, WANG SHUYING, ZHENG YIBIN, REN ZHIGANG, YANG JIE
• Format: Patent
• Language: Chinese; English
• Published: 17.01.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention discloses a microarchitecture searching method and device for a deep convolutional neural network, is applied to a power transmission line hidden danger analysis scene or a substation defect detection scene, and belongs to the technical field of neural network architecture searching. The to-be-detected data comprises a power transmission line hidden danger analysis data set or substation defect detection data; calculating a neural network weight and a neural network architecture parameter to obtain an optimal architecture of the neural network; the maximum coding rate reduction of the model intermediate layer features irrelevant to the task is adopted as a network architecture performance standard for searching; the maximum coding rate reduction of the time domain feature and the frequency domain feature of the model intermediate layer related to the task is adopted as a network architecture performance standard for searching. According to the invention, acceleration and compression of the deep