Improving Axial-Attention Network Classification via Cross-Channel Weight Sharing

TL;DR

This paper explores replacing layers in Axial Attention networks with hypercomplex-inspired variants, leading to improved image classification accuracy on ImageNet300k across multiple network architectures.

Contribution

It introduces a method of integrating hypercomplex-inspired layers into Axial Attention networks, demonstrating consistent accuracy improvements across various network components.

Findings

Improved accuracy on ImageNet300k with various modifications

Hypercomplex variants enhance representational coherence

Technique is broadly applicable to different network parts

Abstract

In recent years, hypercomplex-inspired neural networks (HCNNs) have been used to improve deep learning architectures due to their ability to enable channel-based weight sharing, treat colors as a single entity, and improve representational coherence within the layers. The work described herein studies the effect of replacing existing layers in an Axial Attention network with their representationally coherent variants to assess the effect on image classification. We experiment with the stem of the network, the bottleneck layers, and the fully connected backend, by replacing them with representationally coherent variants. These various modifications lead to novel architectures which all yield improved accuracy performance on the ImageNet300k classification dataset. Our baseline networks for comparison were the original real-valued ResNet, the original quaternion-valued ResNet, and the Axial Attention ResNet. Since improvement was observed regardless of which part of the network was modified, there is a promise that this technique may be generally useful in improving classification accuracy for a large class of networks.