Towards Efficient and Accurate Spiking Neural Networks via Adaptive Bit Allocation

2025-06-30

AI Summary

Multi-bit spiking neural networks (SNNs) suffer from rapidly escalating memory and computational overhead as bit-width increases, leading to a deteriorated energy–accuracy trade-off. To address this, we propose the first learnable framework for joint optimization of temporal length and bit-width, featuring an enhanced differentiable spiking neuron model and a gradient-driven step-size update mechanism that enables layer-wise fine-grained adaptive bit allocation. Leveraging quantization-aware modeling and end-to-end differentiable training, our approach effectively mitigates quantization error and gradient mismatch. Extensive experiments on CIFAR, ImageNet, and DVS datasets demonstrate significant improvements: SEWResNet-34 achieves a 2.69% accuracy gain on ImageNet while reducing the bit budget by 4.16×, substantially enhancing both energy efficiency and accuracy of SNNs.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Multi-bit spiking neural networks (SNNs) have recently become a heated research spot, pursuing energy-efficient and high-accurate AI. However, with more bits involved, the associated memory and computation demands escalate to the point where the performance improvements become disproportionate. Based on the insight that different layers demonstrate different importance and extra bits could be wasted and interfering, this paper presents an adaptive bit allocation strategy for direct-trained SNNs, achieving fine-grained layer-wise allocation of memory and computation resources. Thus, SNN's efficiency and accuracy can be improved. Specifically, we parametrize the temporal lengths and the bit widths of weights and spikes, and make them learnable and controllable through gradients. To address the challenges caused by changeable bit widths and temporal lengths, we propose the refined spiking neuron, which can handle different temporal lengths, enable the derivation of gradients for temporal lengths, and suit spike quantization better. In addition, we theoretically formulate the step-size mismatch problem of learnable bit widths, which may incur severe quantization errors to SNN, and accordingly propose the step-size renewal mechanism to alleviate this issue. Experiments on various datasets, including the static CIFAR and ImageNet and the dynamic CIFAR-DVS and DVS-GESTURE, demonstrate that our methods can reduce the overall memory and computation cost while achieving higher accuracy. Particularly, our SEWResNet-34 can achieve a 2.69% accuracy gain and 4.16$ imes$ lower bit budgets over the advanced baseline work on ImageNet. This work will be fully open-sourced.

Problem

• Optimizing bit allocation in SNNs for efficiency and accuracy
• Reducing memory and computation costs in multi-bit SNNs
• Addressing step-size mismatch in learnable bit widths

Innovation

• Adaptive bit allocation for SNN efficiency
• Refined spiking neuron handles variable lengths
• Step-size renewal reduces quantization errors