CHAMP: Coherent Hardware-Aware Magnitude Pruning of Integrated Photonic Neural Networks

TL;DR

This paper introduces CHAMP, a hardware-aware magnitude pruning method for coherent photonic neural networks that significantly reduces parameters and power consumption with minimal accuracy loss.

Contribution

The paper presents a novel pruning technique tailored for photonic neural networks, achieving high sparsity and power efficiency while maintaining accuracy.

Findings

Prunes 99.45% of network parameters.

Reduces static power consumption by 98.23%.

Maintains negligible accuracy loss.

Abstract

We propose a novel hardware-aware magnitude pruning technique for coherent photonic neural networks. The proposed technique can prune 99.45% of network parameters and reduce the static power consumption by 98.23% with a negligible accuracy loss.