Tensorizing Generative Adversarial Nets

Generative Adversarial Network (GAN) and its variants exhibit state-of-the-art performance in the class of generative models. To capture higher-dimensional distributions, the common learning procedure requires high computational complexity and a large number of parameters. The problem of employing such massive framework arises when deploying it on a platform with limited computational power such as mobile phones. In this paper, we present a new generative adversarial framework by representing each layer as a tensor structure connected by multilinear operations, aiming to reduce the number of model parameters by a large factor while preserving the generative performance and sample quality. To learn the model, we employ an efficient algorithm which alternatively optimizes both discriminator and generator. Experimental outcomes demonstrate that our model can achieve high compression rate for model parameters up to $35$ times when compared to the original GAN for MNIST dataset.

Tensor Regression Networks with various Low-Rank Tensor Approximations

Tensor regression networks achieve high rate of compression of model parameters in multilayer perceptrons (MLP) while having slight impact on performances. Tensor regression layer imposes low-rank constraints on the tensor regression layer which replaces the flattening operation of traditional MLP. We investigate tensor regression networks using various low-rank tensor approximations, aiming to leverage the multi-modal structure of high dimensional data by enforcing efficient low-rank constraints. We provide a theoretical analysis giving insights on the choice of the rank parameters. We evaluated performance of proposed model with state-of-the-art deep convolutional models. For CIFAR-10 dataset, we achieved the compression rate of 0.018 with the sacrifice of accuracy less than 1%.