Neural network architecture automatic search method and system for liquid state machine, and medium

• Main Authors: WANG LEI, ZHANG JIANFENG, ZHANG YING, LI NAN, TIAN SHUO, QU LIANHUA, GUO SHASHA, HU KAI, KANG ZIYANG, XU WEIXIA
• Format: Patent
• Language: Chinese; English
• Published: 03.11.2020
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention discloses a neural network architecture automatic search method and system for a liquid state machine, and a medium, and the method comprises the steps: converting an inputted proxy dataset into a pulse mode, and randomly generating an initial parent LSM architecture; by a SNN simulator, according to the proxy data set and the parent LSM architecture of the pulse mode, extracting and sending the state of the liquid after data set recognition to a machine learning classifier for training and testing, and acquiring the classification precision of the liquid through the machine learning classifier; judging whether the NAS exploration frequency reaches a preset threshold value or not, and if the NAS exploration frequency does not reach the preset threshold value, performing NASexploration on the current parent LSM architecture to obtain a child LSM architecture as a new parent LSM architecture for re-iteration; otherwise, carrying out ending and exiting. The architecture and parameter design space o