Frequency Specific Effects of Oscillatory Inputs on Timing and Chaotic Time-Series Learning in Spiking Reservoir Computing

Jihoon Park, Yuji Kawai, Minoru Asada

Abstract: The brain performs cognitive and motor functions with precise timing that may be driven by neural oscillatory activities such as hippocampal theta waves. Inspired by these biological mechanisms, oscillation-driven reservoir computing is a framework that improves the performance of reservoir computing in temporal processing including a timing task. In this study, we investigate the computational roles of frequency bands of oscillatory inputs for the timing task and the prediction task of a chaotic time series using oscillation-driven spiking reservoir computing (ODSRC) based on the biologically plausible Izhikevich neuron model. The results showed that the ODSRC effectively performed both tasks for durations exceeding 30 s. Moreover, our experiments across various frequency bands showed that low-frequency oscillatory inputs enhance the reproduction of target signals, whereas high-frequency oscillations improve generalization performance. This study promises ODSRC to be an energy-efficient and biologically inspired approach to improving long-term temporal processing in neurocomputing systems.