PAC-Bayesian Contrastive Unsupervised Representation Learning

TL;DR

This paper extends theoretical understanding of contrastive unsupervised representation learning (CURL) by deriving PAC-Bayesian generalisation bounds, leading to a new algorithm that performs competitively on real datasets.

Contribution

It introduces PAC-Bayesian generalisation bounds for CURL and develops a new representation learning algorithm based on these bounds.

Findings

The new algorithm achieves competitive accuracy on real datasets.

The bounds are non-vacuous and provide meaningful generalisation guarantees.

Empirical results support the theoretical analysis.

Abstract

Contrastive unsupervised representation learning (CURL) is the state-of-the-art technique to learn representations (as a set of features) from unlabelled data. While CURL has collected several empirical successes recently, theoretical understanding of its performance was still missing. In a recent work, Arora et al. (2019) provide the first generalisation bounds for CURL, relying on a Rademacher complexity. We extend their framework to the flexible PAC-Bayes setting, allowing us to deal with the non-iid setting. We present PAC-Bayesian generalisation bounds for CURL, which are then used to derive a new representation learning algorithm. Numerical experiments on real-life datasets illustrate that our algorithm achieves competitive accuracy, and yields non-vacuous generalisation bounds.