A study on reservoir computing using a single P-HCNM with self-feedback

カテゴリ：国際会議

論文No：D6

グループ名：【C】AVIC2025

発行日：2025/10/20

著者名：Ryo Ono (Nihon University), Takeru Yonekawa (Nihon University), Takuto Yamaguchi (Nihon University), Katsutoshi Saeki (Nihon University)

キーワード：Physical reservoir computing,Pulse-type hardware chaotic neuron model,Self-feedback,Temporal XOR task,Delay task

要約(英語)：Reservoir computing (RC), a computational framework derived from recurrent neural networks (RNNs), has attracted significant attention for its ease of implementation in physical systems. Because it is primarily due to the characteristic feature of the RC in which the connection weights of the reservoir layer are fixed, reducing the complexity of learning. In previous study, we investigated RC using a neural network of Pulse-type Hardware Chaotic Neuron Models (P-HCNMs), which are electronic circuit models that mimic biological neurons. However, the RC using electronic circuit models which mimic biological neurons have not yet been implemented in hardware. In this study, we implemented a single P-HCNM with self-feedback using discrete components and constructed a hardware-based reservoir computing system. To evaluate its performance, we have conducted a temporal XOR task and a Delay task, a typical benchmark in the RC. The system has achieved a low bit error rate of 4.7% and a high memory capacity of 1.07, demonstrating that the RC can be realized using a single PHCNM with feedback.

本誌掲載ページ：89-92p

原稿種別：英語

PDFファイルサイズ：891Kバイト