Fast Clifford Neural Layers

TL;DR

This paper introduces optimized Clifford neural layers for PDE modeling, achieving 30% faster inference on CPU compared to standard implementations, and provides open-source code for practical use.

Contribution

It presents an optimized implementation of Clifford convolutional and multivector activation layers for CPU inference, enhancing performance in PDE modeling applications.

Findings

30% faster inference on CPU compared to standard PyTorch

Effective optimization for large data and network sizes

Open-source code available for community use

Abstract

Clifford Neural Layers improve PDE modeling by introducing Clifford Algebra into neural networks. In this project we focus on optimizing the inference of 2/3D Clifford convolutional layers and multivector activation layers for one core CPU performance. Overall, by testing on a real network block involving Clifford convolutional layers and multivector activation layers, we observe that our implementation is 30% faster than standard PyTorch implementation in relatively large data + network size (>L2 cache). We open source our code base at https://github.com/egretwAlker/c-opt-clifford-layers