Mental-State Estimation Using Multi-FPGA SNN Based on Liquid State Machine
Haruto Ota, Ryuji Nagazawa, Kien Nguyen, Nobuyoshi Komuro and Hiroo Sekiya
2026 RISP International Workshop on Nonlinear Circuits, Communications and Signal Processing, Feb., 2026. [pdf document]

<Abstract>

This paper proposes a multi-FPGA SNN based on a liquid state machine (LSM) for estimating emotions from environmental data. The proposed method introduces an independent two-axis estimation model for valence and arousal based on Russell’s circumplex model. Furthermore, a time window aligned with the neuronal time constant is applied to the gradual variations in environmental data to enhance the transient response characteristics of the reservoir. Experimental results demonstrate that the proposed system achieved an accuracy of 90.1 % in evaluations considering margins.