SAFformer:Improving Spiking Transformer via Active Predictive Filtering

TL;DR

SAFformer introduces an active predictive filtering approach to Spiking Transformers, inspired by brain mechanisms, significantly enhancing accuracy and efficiency on visual tasks.

Contribution

It proposes a novel active filtering paradigm for Spiking Transformers, improving focus on salient features and reducing computational overhead.

Findings

Achieves state-of-the-art results on CIFAR-10/100 and CIFAR10-DVS datasets.

On ImageNet-1K, reaches 80.50% Top-1 accuracy with low energy consumption.

Demonstrates a new balance between accuracy and energy efficiency in Spiking Transformers.

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 80.50% Top-1 accuracy with only 26.58M parameters and an energy consumption of 5.88 mJ, demonstrating an exceptional balance between accuracy and efficiency.