A study on feedback architecture for STDP learning in hierarchical neural networks using P-HCNM

カテゴリ：国際会議

論文No：D2

グループ名：【C】AVIC2025

発行日：2025/10/20

著者名：Fuya Imamura (Nihon University), Yoshiki Sasaki (Nihon University)

キーワード：Spiking neuron,VLSI,Analog electrical circuits,Hierarchical neural network,Spice simulation

要約(英語)：In recent years, research has been conducted on constructing networks capable of brain-inspired computing, which emulates the biological processing and recall abilities of external information. In the field of physiology, Spike Timing Dependent Plasticity (STDP)—a training rule that adjusts synaptic weights based on the timing difference between the firing of preceding and post neurons—has attracted significant attention. However, in previous studies, training signals were applied to both the preceding and post synaptic neurons to forcibly induce firing with a time difference. As a result, training was performed without relying on the natural timing of preceding and post synaptic firing. In this paper, we examine the architecture of a hierarchical neural network that enables training using an asymmetric STDP circuit, without the need to manually input timing differences to the preceding and post synaptic P-HCNMs in the hardware structure. As a result, we demonstrated that by introducing excitatory synaptic feedback connections from the preceding to post synaptic and from the post to preceding synaptic stages, it is possible to learn and recall 9 x 9 character image data, and perform classification of the input data.

本誌掲載ページ：73-76p

原稿種別：英語

PDFファイルサイズ：730Kバイト