Proceedings

SAFformer: Improving Spiking Transformer via Active Predictive Filtering

Zequan Xie, Weiming Zeng, Yunhua Chen, Sichang Lin, Tongyang Chen, Jinsheng Xiao

Download BibTeX Cite

Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence. Main Track. Pages 1902-1910.

https://doi.org/10.24963/ijcai.2026/212

Keywords

  • Humans and AI: Cognitive systems
  • Computer Vision: Efficiency and Optimization
  • Humans and AI: Cognitive modeling
  • Machine Learning: Attention models

Abstract

Spiking Neural Networks (SNNs) offer notable advantages in biological plausibility and energy efficiency, making them promising candidates for building low-power Transformers. However, existing Spiking Transformers largely adhere to a passive reactive paradigm, which struggles to focus on task-relevant information and incurs substantial computational overhead when processing redundant visual data. To overcome this fundamental yet underexplored limitation, we propose SAFformer, a novel Spiking Transformer architecture based on an active predictive filtering paradigm. Inspired by the brain’s predictive coding mechanism, SAFformer actively suppresses predictable signals and focuses on salient visual features. Extensive experiments show that SAFformer establishes new state-of-the-art performance on CIFAR-10/100 and CIFAR10-DVS. Remarkably, on ImageNet-1K, it achieves 80.44% Top-1 accuracy with only 26.58M parameters and an energy consumption of 5.88 mJ, demonstrating an exceptional balance between accuracy and efficiency.

Zequan Xie, Weiming Zeng, Yunhua Chen, Sichang Lin, Tongyang Chen, and Jinsheng Xiao. 2026. SAFformer: Improving Spiking Transformer via Active Predictive Filtering. In Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26. IJCAI, 1902–1910. https://doi.org/10.24963/ijcai.2026/212
Welcome Future Conferences AI Journal Division IJCAI Proceedings About IJCAI IJCAI Awards