Building Efficient Deep Neural Networks with Unitary Group Convolutions
This work addresses efficiency improvements in CNN architectures for computer vision applications, presenting a novel method that unifies and enhances existing techniques.
The authors tackled the problem of building efficient deep neural networks by proposing unitary group convolutions (UGConvs), which combine group convolutions with unitary transforms to learn richer representations, and introduced HadaNet using Hadamard transforms, achieving similar accuracy to circulant networks with lower computation complexity and better accuracy than ShuffleNets with the same parameters and floating-point multiplies.
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.