Building Efficient Deep Neural Networks with Unitary Group Convolutions

TL;DR

This paper introduces unitary group convolutions (UGConvs) that enhance CNNs by combining group convolutions with unitary transforms, unifying existing techniques and improving accuracy and efficiency.

Contribution

The paper proposes UGConvs as a new building block for CNNs, unifying channel shuffling and block-circulant networks, and introduces HadaNet with Hadamard transforms for improved performance.

Findings

Dense unitary transforms outperform channel shuffling in accuracy.

Different dense transforms have similar accuracy performance.

HadaNet achieves comparable accuracy to circulant networks with lower complexity.

Abstract

We propose unitary group convolutions (UGConvs), a building block for CNNs which compose a group convolution with unitary transforms in feature space to learn a richer set of representations than group convolution alone. UGConvs generalize two disparate ideas in CNN architecture, channel shuffling (i.e. ShuffleNet) and block-circulant networks (i.e. CirCNN), and provide unifying insights that lead to a deeper understanding of each technique. We experimentally demonstrate that dense unitary transforms can outperform channel shuffling in DNN accuracy. On the other hand, different dense transforms exhibit comparable accuracy performance. Based on these observations we propose HadaNet, a UGConv network using Hadamard transforms. HadaNets achieve similar accuracy to circulant networks with lower computation complexity, and better accuracy than ShuffleNets with the same number of parameters and floating-point multiplies.