Elementary Cellular Automata for Reservoir Computing with Modest Resources
Kazuhiro Yokota
Ion Technology Center Co., Ltd
Hirakata, Osaka, 573-0128, Japan
yokota@iontc.co.jp
Abstract
Several elementary cellular automaton (ECA) rules have been found useful for reservoir computing (RC). In this paper, we investigate how effectively ECA rules operate with reduced computational resources. We examine successful rules with the 5-bit memory task and nonlinear autoregressive moving-average (NARMA) benchmarks. Our model is a slightly modified version of one previously reported that is optimized for modest resources. We find that the features produced by a cellular automaton (CA) at each timestep vary greatly in their contributions to the computation. The performance on the NARMA task is improved, especially in a rule from Wolfram’s class 2. The result demonstrates the feasibility of implementing RC with modest resources, such as those in embedded systems, and helps elucidate the role of cellular automata in RC.
Keywords: reservoir computing; elementary cellular automaton
Cite this publication as:
K. Yokota, “Elementary Cellular Automata for Reservoir Computing with Modest Resources,” Complex Systems, 35(2), 2026 pp. 161–180.
https://doi.org/10.25088/ComplexSystems.35.2.161
