Self-Supervised Learning by Estimating Twin Class Distributions

TL;DR

TWIST is a self-supervised learning method that uses twin class distributions and mutual information maximization to learn effective representations without collapse, outperforming existing methods especially in semi-supervised settings.

Contribution

The paper introduces TWIST, a novel self-supervised learning approach that avoids collapse without complex tricks by maximizing mutual information between inputs and predictions.

Findings

TWIST outperforms state-of-the-art methods on various tasks.

Achieves 61.2% top-1 accuracy with 1% labeled ImageNet data.

Effectively prevents collapse without asymmetric networks or stop-gradient.

Abstract

We present TWIST, a simple and theoretically explainable self-supervised representation learning method by classifying large-scale unlabeled datasets in an end-to-end way. We employ a siamese network terminated by a softmax operation to produce twin class distributions of two augmented images. Without supervision, we enforce the class distributions of different augmentations to be consistent. However, simply minimizing the divergence between augmentations will cause collapsed solutions, i.e., outputting the same class probability distribution for all images. In this case, no information about the input image is left. To solve this problem, we propose to maximize the mutual information between the input and the class predictions. Specifically, we minimize the entropy of the distribution for each sample to make the class prediction for each sample assertive and maximize the entropy of the mean distribution to make the predictions of different samples diverse. In this way, TWIST can naturally avoid the collapsed solutions without specific designs such as asymmetric network, stop-gradient operation, or momentum encoder. As a result, TWIST outperforms state-of-the-art methods on a wide range of tasks. Especially, TWIST performs surprisingly well on semi-supervised learning, achieving 61.2% top-1 accuracy with 1% ImageNet labels using a ResNet-50 as backbone, surpassing previous best results by an absolute improvement of 6.2%. Codes and pre-trained models are given on: https://github.com/bytedance/TWIST