Plug-and-Play Spiking Operators: Breaking the Nonlinearity Bottleneck in Spiking Transformers
A new framework called Plug-and-Play Spiking Operators has been proposed to enhance the performance of spiking transformers. This framework addresses the limitations of current ANN-to-SNN conversion methods by implementing spike-friendly approximations for nonlinear operations. Experiments indicate that the proposed method incurs minimal accuracy loss while integrating into existing pipelines.
- ▪The framework decomposes nonlinear computations into three primitives: division, exponentiation, and L2 norms.
- ▪It uses population computation with leaky integrate-and-fire neuron groups to avoid floating-point arithmetic.
- ▪Selective replacement of nonlinear operators results in less than a 1% accuracy drop across various tasks.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Machine Learning arXiv:2605.20289 (cs) [Submitted on 19 May 2026] Title:Plug-and-Play Spiking Operators: Breaking the Nonlinearity Bottleneck in Spiking Transformers Authors:Xinzhe Yuan (1), Xiang Peng (1), Bin Gu (2), Huan Xiong (1) ((1) IASM, Harbin Institute of Technology, (2) School of Artificial Intelligence, Jilin University) View a PDF of the paper titled Plug-and-Play Spiking Operators: Breaking the Nonlinearity Bottleneck in Spiking Transformers, by Xinzhe Yuan (1) and 6 other authors View PDF HTML (experimental) Abstract:ANN-to-SNN conversion offers a practical, training-free route to spiking large language models.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
