Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

TLDR

Convolutional neural networks have dominated supervised computer vision, while unsupervised CNN approaches remain comparatively underexplored. This study aims to bridge that gap by exploring unsupervised learning with CNNs. The authors propose deep convolutional generative adversarial networks (DCGANs) with specific architectural constraints as a promising unsupervised CNN framework. Experiments on multiple image datasets reveal that DCGANs learn hierarchical representations from object parts to scenes, and the extracted features transfer effectively to new tasks.

Abstract

In recent years, supervised learning with convolutional networks (CNNs) has seen huge adoption in computer vision applications. Comparatively, unsupervised learning with CNNs has received less attention. In this work we hope to help bridge the gap between the success of CNNs for supervised learning and unsupervised learning. We introduce a class of CNNs called deep convolutional generative adversarial networks (DCGANs), that have certain architectural constraints, and demonstrate that they are a strong candidate for unsupervised learning. Training on various image datasets, we show convincing evidence that our deep convolutional adversarial pair learns a hierarchy of representations from object parts to scenes in both the generator and discriminator. Additionally, we use the learned features for novel tasks - demonstrating their applicability as general image representations.