TCL: an ANN-to-SNN Conversion with Trainable Clipping Layers

TL;DR

This paper introduces TCL, a method for converting ANNs to SNNs with trainable clipping layers, significantly reducing the accuracy-latency trade-off and achieving high accuracy on ImageNet with moderate latency.

Contribution

TCL enables efficient ANN-to-SNN conversion by incorporating trainable clipping layers, improving accuracy-latency trade-offs on large datasets like ImageNet.

Findings

Achieved 73.87% accuracy with VGG-16 on ImageNet.

Achieved 70.37% accuracy with ResNet-34 on ImageNet.

Moderate latency of 250 cycles in SNNs.

Abstract

Spiking-neural-networks (SNNs) are promising at edge devices since the event-driven operations of SNNs provides significantly lower power compared to analog-neural-networks (ANNs). Although it is difficult to efficiently train SNNs, many techniques to convert trained ANNs to SNNs have been developed. However, after the conversion, a trade-off relation between accuracy and latency exists in SNNs, causing considerable latency in large size datasets such as ImageNet. We present a technique, named as TCL, to alleviate the trade-off problem, enabling the accuracy of 73.87% (VGG-16) and 70.37% (ResNet-34) for ImageNet with the moderate latency of 250 cycles in SNNs.