Optimising Event-Driven Spiking Neural Network with Regularisation and Cutoff

2023-01-23

arXiv.org

AI Summary

To address the low inference efficiency and high temporal overhead of Spiking Neural Networks (SNNs), this paper proposes a dynamic inference termination mechanism that adaptively halts event-driven inference while preserving accuracy. Our key contributions are: (1) the first Top-K dynamic termination strategy, which determines inference completion in real time based on neuronal activation strength; (2) termination-aware regularization, explicitly incorporating termination behavior into the training objective to ensure training–inference consistency; and (3) full compatibility with both ANN-to-SNN conversion and direct SNN training paradigms. Evaluated on CIFAR-10 (frame-based data), our method reduces average simulation timesteps by 2.26×; on CIFAR10-DVS (event-based data), it achieves a 1.79× reduction. Accuracy degradation remains below 0.3%, yielding substantial improvements in energy efficiency.

Abstract

Spiking neural network (SNN), as the next generation of artificial neural network (ANN), offer a closer mimicry of natural neural networks and hold promise for significant improvements in computational efficiency. However, the current SNN is trained to infer over a fixed duration, overlooking the potential of dynamic inference in SNN. In this paper, we strengthen the marriage between SNN and event-driven processing with a proposal to consider a cutoff in SNN, which can terminate SNN anytime during inference to achieve efficient inference. Two novel optimisation techniques are presented to achieve inference efficient SNN: a Top-K cutoff and a regularisation.The proposed regularisation influences the training process, optimising SNN for the cutoff, while the Top-K cutoff technique optimises the inference phase. We conduct an extensive set of experiments on multiple benchmark frame-based datasets, such asCIFAR10/100, Tiny-ImageNet, and event-based datasets, including CIFAR10-DVS, N-Caltech101 and DVS128 Gesture. The experimental results demonstrate the effectiveness of our techniques in both ANN-to-SNN conversion and direct training, enabling SNNs to require 1.76 to 2.76x fewer timesteps for CIFAR-10, while achieving 1.64 to 1.95x fewer timesteps across all event-based datasets, with near-zero accuracy loss. These findings affirms the compatibility and potential benefits of our techniques in enhancing accuracy and reducing inference latency when integrated with existing methods. Code available: https://github.com/Dengyu-Wu/SNNCutoff

Problem

• Spiking Neural Networks
• Efficiency Optimization
• Accuracy Preservation

Innovation

• Top-K cutoff
• Regularization technique
• Spiking Neural Networks (SNN) efficiency enhancement