Reconstructing Spiking Neural Networks Using a Single Neuron with Autapses

2026-03-25

AI Summary

This work proposes a time-delayed autaptic spiking neural network (TDA-SNN) to address the high communication overhead and state storage costs inherent in conventional multi-layer dense SNN architectures. By incorporating time-delayed autaptic connections within a single leaky integrate-and-fire neuron, TDA-SNN unifies reservoir computing, multilayer perceptron, and convolution-like operations through prototype learning and internal temporal state reuse. This approach represents the first integration of multiple SNN computational paradigms within a single neuron, substantially reducing both neuron count and state memory requirements while enhancing the information capacity per neuron. Evaluated on sequence modeling, event-based, and image classification benchmarks, TDA-SNN achieves competitive performance, demonstrating its favorable space–time trade-off and offering a highly compact computational unit for brain-inspired computing.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Deep Neural Architectures and Foundation Models
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Social Networks and Social Media: Computational social science

Abstract

Spiking neural networks (SNNs) are promising for neuromorphic computing, but high-performing models still rely on dense multilayer architectures with substantial communication and state-storage costs. Inspired by autapses, we propose time-delayed autapse SNN (TDA-SNN), a framework that reconstructs SNNs with a single leaky integrate-and-fire neuron and a prototype-learning-based training strategy. By reorganizing internal temporal states, TDA-SNN can realize reservoir, multilayer perceptron, and convolution-like spiking architectures within a unified framework. Experiments on sequential, event-based, and image benchmarks show competitive performance in reservoir and MLP settings, while convolutional results reveal a clear space--time trade-off. Compared with standard SNNs, TDA-SNN greatly reduces neuron count and state memory while increasing per-neuron information capacity, at the cost of additional temporal latency in extreme single-neuron settings. These findings highlight the potential of temporally multiplexed single-neuron models as compact computational units for brain-inspired computing.

Problem

• Spiking Neural Networks
• Neuromorphic Computing
• Model Compression
• Autapses
• Temporal Multiplexing

Innovation

• spiking neural networks
• autapse
• single-neuron computing
• temporal multiplexing
• neuromorphic computing