Discriminative Unsupervised Feature Learning with Convolutional Neural Networks

TLDR

Convolutional neural network training typically requires large labeled datasets. The paper proposes training a CNN solely with unlabeled data. The method trains the network to distinguish surrogate classes created by applying diverse transformations to randomly sampled seed patches. The learned features achieve state‑of‑the‑art unsupervised recognition performance on STL‑10, CIFAR‑10, and Caltech‑101.

Abstract

Current methods for training convolutional neural networks depend on large amounts of labeled samples for supervised training. In this paper we present an approach for training a convolutional neural network using only unlabeled data. We train the network to discriminate between a set of surrogate classes. Each surrogate class is formed by applying a variety of transformations to a randomly sampled 'seed' image patch. We find that this simple feature learning algorithm is surprisingly successful when applied to visual object recognition. The feature representation learned by our algorithm achieves classification results matching or outperforming the current state-of-the-art for unsupervised learning on several popular datasets (STL-10, CIFAR-10, Caltech-101).