Learning Neuron Dynamics within Deep Spiking Neural Networks

2025-10-07

AI Summary

Existing deep spiking neural networks (SNNs) predominantly employ simplified neuron models—such as the leaky integrate-and-fire (LIF) model—that inadequately capture complex temporal dynamics; meanwhile, expressive alternatives suffer from manual hyperparameter tuning and poor scalability. To address this, we propose the Learnable Neuron Model (LNM), which parameterizes the nonlinear integrate-and-fire dynamics via low-order polynomials, enabling end-to-end, data-driven learning of neuron behavior without hand-crafted hyperparameters. LNM is natively compatible with backpropagation and mainstream deep learning frameworks. Equipped with LNM, deep SNNs achieve stable and efficient training. Extensive experiments on static-image benchmarks (CIFAR-10/100, ImageNet) and neuromorphic event-based datasets (CIFAR-10 DVS) demonstrate state-of-the-art performance, significantly enhancing SNNs’ capacity to model temporal structure and improve generalization.

Technology Category

Cognitive Modeling & Cognitive Systems: Neural Spike CodingMachine Learning: Deep Neural Architectures and Foundation ModelsNatural Language Processing: Learning & Optimization for NLP

Application Category

Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphsSearch and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved informationEconomics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Spiking Neural Networks (SNNs) offer a promising energy-efficient alternative to Artificial Neural Networks (ANNs) by utilizing sparse and asynchronous processing through discrete spike-based computation. However, the performance of deep SNNs remains limited by their reliance on simple neuron models, such as the Leaky Integrate-and-Fire (LIF) model, which cannot capture rich temporal dynamics. While more expressive neuron models exist, they require careful manual tuning of hyperparameters and are difficult to scale effectively. This difficulty is evident in the lack of successful implementations of complex neuron models in high-performance deep SNNs. In this work, we address this limitation by introducing Learnable Neuron Models (LNMs). LNMs are a general, parametric formulation for non-linear integrate-and-fire dynamics that learn neuron dynamics during training. By learning neuron dynamics directly from data, LNMs enhance the performance of deep SNNs. We instantiate LNMs using low-degree polynomial parameterizations, enabling efficient and stable training. We demonstrate state-of-the-art performance in a variety of datasets, including CIFAR-10, CIFAR-100, ImageNet, and CIFAR-10 DVS. LNMs offer a promising path toward more scalable and high-performing spiking architectures.

Problem

• Deep SNNs underperform due to simplistic neuron models
• Complex neuron models require difficult manual hyperparameter tuning
• Lack of scalable complex neuron implementations in deep SNNs

Innovation

• Learnable Neuron Models learn dynamics from data
• Parametric formulation enables stable training process
• Polynomial parameterizations achieve state-of-the-art performance