SAFformer:Improving Spiking Transformer via Active Predictive Filtering

SAFformer is a novel Spiking Transformer architecture that utilizes an active predictive filtering approach to enhance focus on task-relevant information while minimizing computational overhead. It achieves state-of-the-art performance on CIFAR-10/100 and CIFAR10-DVS, and on ImageNet-1K, it reaches 80.44% Top-1 accuracy with only 26.58 million parameters and an energy consumption of 5.88 mJ. This architecture offers a promising solution for practitioners aiming to build low-power and efficient models in the domain of Spiking Neural Networks.