Hybrid Generative-Contrastive Representation Learning

TL;DR

This paper proposes a hybrid training scheme combining contrastive and generative losses in a transformer-based model to learn representations that are both highly discriminative and robust, validated across multiple tasks.

Contribution

It introduces a novel hybrid training approach that leverages the strengths of both contrastive and generative learning in a single model.

Findings

The hybrid model achieves superior discriminative performance.

The model maintains strong generative capabilities.

It improves robustness to out-of-distribution data.

Abstract

Unsupervised representation learning has recently received lots of interest due to its powerful generalizability through effectively leveraging large-scale unlabeled data. There are two prevalent approaches for this, contrastive learning and generative pre-training, where the former learns representations from instance-wise discrimination tasks and the latter learns them from estimating the likelihood. These seemingly orthogonal approaches have their own strengths and weaknesses. Contrastive learning tends to extract semantic information and discards details irrelevant for classifying objects, making the representations effective for discriminative tasks while degrading robustness to out-of-distribution data. On the other hand, the generative pre-training directly estimates the data distribution, so the representations tend to be robust but not optimal for discriminative tasks. In this paper, we show that we could achieve the best of both worlds by a hybrid training scheme. Specifically, we demonstrated that a transformer-based encoder-decoder architecture trained with both contrastive and generative losses can learn highly discriminative and robust representations without hurting the generative performance. We extensively validate our approach on various tasks.