Can Deep Neural Networks be Converted to Ultra Low-Latency Spiking Neural Networks?

TL;DR

This paper introduces a new training method for converting deep neural networks into ultra low-latency spiking neural networks, significantly reducing energy consumption and inference time while maintaining accuracy.

Contribution

The authors develop an accurate distribution-aware training algorithm that enables ultra low-latency SNNs with high sparsity, outperforming existing conversion strategies.

Findings

Achieved 64.19% top-1 accuracy with 2 time steps on CIFAR-100.

Reduced compute energy by approximately 159.2x compared to DNNs.

Inference speed improved by 2.5-8x over other SOTA SNN models.

Abstract

Spiking neural networks (SNNs), that operate via binary spikes distributed over time, have emerged as a promising energy efficient ML paradigm for resource-constrained devices. However, the current state-of-the-art (SOTA) SNNs require multiple time steps for acceptable inference accuracy, increasing spiking activity and, consequently, energy consumption. SOTA training strategies for SNNs involve conversion from a non-spiking deep neural network (DNN). In this paper, we determine that SOTA conversion strategies cannot yield ultra low latency because they incorrectly assume that the DNN and SNN pre-activation values are uniformly distributed. We propose a new training algorithm that accurately captures these distributions, minimizing the error between the DNN and converted SNN. The resulting SNNs have ultra low latency and high activation sparsity, yielding significant improvements in compute efficiency. In particular, we evaluate our framework on image recognition tasks from CIFAR-10 and CIFAR-100 datasets on several VGG and ResNet architectures. We obtain top-1 accuracy of 64.19% with only 2 time steps on the CIFAR-100 dataset with ~159.2x lower compute energy compared to an iso-architecture standard DNN. Compared to other SOTA SNN models, our models perform inference 2.5-8x faster (i.e., with fewer time steps).