Unsupervised Representation Learning via Neural Activation Coding

TL;DR

Neural activation coding (NAC) is a novel unsupervised learning method that maximizes the mutual information between encoder activations and data, enhancing representation expressivity and improving downstream task performance.

Contribution

NAC introduces a new approach that increases nonlinear expressivity of deep encoders by maximizing mutual information, learning both continuous and discrete data representations.

Findings

NAC improves linear classification accuracy on CIFAR-10 and ImageNet-1K.

NAC enhances nearest neighbor retrieval performance on CIFAR-10 and FLICKR-25K.

NAC outperforms or matches recent baselines like SimCLR and DistillHash.

Abstract

We present neural activation coding (NAC) as a novel approach for learning deep representations from unlabeled data for downstream applications. We argue that the deep encoder should maximize its nonlinear expressivity on the data for downstream predictors to take full advantage of its representation power. To this end, NAC maximizes the mutual information between activation patterns of the encoder and the data over a noisy communication channel. We show that learning for a noise-robust activation code increases the number of distinct linear regions of ReLU encoders, hence the maximum nonlinear expressivity. More interestingly, NAC learns both continuous and discrete representations of data, which we respectively evaluate on two downstream tasks: (i) linear classification on CIFAR-10 and ImageNet-1K and (ii) nearest neighbor retrieval on CIFAR-10 and FLICKR-25K. Empirical results show that NAC attains better or comparable performance on both tasks over recent baselines including SimCLR and DistillHash. In addition, NAC pretraining provides significant benefits to the training of deep generative models. Our code is available at https://github.com/yookoon/nac.