Scholarlink

<p>Spiking Neural Networks (SNNs) offer an energy-efficient alternative to conventional artificial neural networks (ANNs). Although ANN-to-SNN conversion has emerged as a promising direction for developing high-performance spike-driven models with reduced training complexity, it still faces the challenge of maintaining energy efficiency in converted SNNs. In this paper, we address these limitations by introducing a dynamic spiking token mixer, inspired by the strong information redundancy present in the spike self-attention mechanism. Our approach effectively replicates the selective processing capabilities of self-attention through dynamic token masking (DynMask), with layer-specific masking ratios customized according to both spatial and temporal significance. Comprehensive results establish DynMask as a practical step toward efficient deep learning systems. Specifically, DynMask achieves performance gains of up to +3.23% on ImageNet-1K while narrowing the accuracy gaps to as low as +0.02% compared to ANNs, with simultaneous energy consumption reductions of up to 44<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation>. Moreover, our approach successfully extends to complex vision tasks that remain largely unexplored in SNN literature, including COCO detection and ADE20K segmentation. Code: <a href="https://github.com/bic-L/Masked-Spiking-Transformer">https://github.com/bic-L/Masked-Spiking-Transformer</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dynamic Token Masking in Spiking Neural Network

  • 期刊论文

  • 发表日期：

    2026年5月10日

查看全文
  • Yuetong Fang,
  • Ziqing Wang,
  • Deming Zhou,
  • Hongwei Ren,
  • Shibo Zhou,
  • Renjing Xu

摘要

Spiking Neural Networks (SNNs) offer an energy-efficient alternative to conventional artificial neural networks (ANNs). Although ANN-to-SNN conversion has emerged as a promising direction for developing high-performance spike-driven models with reduced training complexity, it still faces the challenge of maintaining energy efficiency in converted SNNs. In this paper, we address these limitations by introducing a dynamic spiking token mixer, inspired by the strong information redundancy present in the spike self-attention mechanism. Our approach effectively replicates the selective processing capabilities of self-attention through dynamic token masking (DynMask), with layer-specific masking ratios customized according to both spatial and temporal significance. Comprehensive results establish DynMask as a practical step toward efficient deep learning systems. Specifically, DynMask achieves performance gains of up to +3.23% on ImageNet-1K while narrowing the accuracy gaps to as low as +0.02% compared to ANNs, with simultaneous energy consumption reductions of up to 44 \(\times \) × . Moreover, our approach successfully extends to complex vision tasks that remain largely unexplored in SNN literature, including COCO detection and ADE20K segmentation. Code: https://github.com/bic-L/Masked-Spiking-Transformer.