Deep convolutional neural networks (CNNs) are successfully used in a number of applications. However, their storage and computational requirements have largely prevented their widespread use on mobile devices. Here we present an effective CNN compression approach in the frequency domain, which focuses not only on smaller weights but on all the weights and their underlying connections. By treating convolutional filters as images, we decompose their representations in the frequency domain as common parts (i.e., cluster centers) shared by other similar filters and their individual private parts (i.e., individual residuals). A large number of low-energy frequency coefficients in both parts can be discarded to produce high compression without significantly compromising accuracy. We relax the computational burden of convolution operations in CNNs by linearly combining the convolution responses of discrete cosine transform (DCT) bases. The compression and speed-up ratios of the proposed algorithm are thoroughly analyzed and evaluated on benchmark image datasets to demonstrate its superiority over state-of-the-art methods.

The technique is fairly natural, since DCTs are routinely used to compress images. However, neural network weight matrices look quite different from images especially in the case of weight decay. They are often sparse, which lends strength to techniques that take advantage of sparsity elementwise rather than sparsity after transformation. This handwaving is backed up by the paper results, which do not exhibit significantly more compression than the P QH baseline. The comparison to only AlexNet and VGG on ImageNet is problematic for judging compression quality, since Inception is significantly smaller than both models. The possibly underconverged non-augmented version on CIFAR-10 is insufficient here, since a better tuned model might be more difficult to compress without losing accuracy.

Deep convolutional neural networks (CNNs) are successfully used in a number of applications. However, their storage and computational requirements have largely prevented their widespread use on mobile devices. Here we present an effective CNN compression approach in the frequency domain, which focuses not only on smaller weights but on all the weights and their underlying connections. By treating convolutional filters as images, we decompose their representations in the frequency domain as common parts (i.e., cluster centers) shared by other similar filters and their individual private parts (i.e., individual residuals). A large number of low-energy frequency coefficients in both parts can be discarded to produce high compression without significantly compromising accuracy.