Presentation Information
[N-2-25]Optimization of inter-reservoir connections in multi-reservoir computing based on spatial light modulator
◎Kaito Kusumi1, Sion Park1, Ziqiang Li2, Gouhei Tanaka2, Tomoki Yamagami1, Atsushi Uchida1 (1. Saitama University, Department of Information and Computer Sciences, Uchida Laboratory, 2. Graduate School of Engineering, Nagoya Institute of Technology)
Keywords:
Reservoir computing,Spatial light modulator,Genetic algorithm,Time-series prediction,Classification
Reservoir computing is a machine learning framework for time-series processing that reduces computational cost by training only output weights. Optical reservoir computing using spatial light modulators has attracted attention for high-speed and energy-efficient information processing. Multi-reservoir computing can improve performance by using multiple interconnected reservoirs. However, genetic algorithm-based optimization of inter-reservoir connections using a spatial light modulator model remains insufficiently investigated. In this study, we optimize inter-reservoir connections in spatial light modulator model-based multi-reservoir computing using a genetic algorithm. The tanh function in the echo state network model and the sin function representing the spatial light modulator model are used as nonlinear functions. Inter-reservoir connections are encoded as individuals, and optimized structures are searched through selection and mutation based on task performance. Performance is evaluated using ECG200 classification and Mackey–Glass prediction tasks. The results show that the optimized structure achieves the best performance for both tasks and both nonlinear functions.