Hybrid Generative-Contrastive Representation Learning
This work addresses the trade-off between discriminative performance and robustness in unsupervised representation learning for machine learning practitioners, offering an incremental improvement by integrating existing methods.
The paper tackles the problem of combining the strengths of contrastive and generative unsupervised representation learning to achieve both discriminative and robust representations, demonstrating that a hybrid training scheme with transformer-based encoder-decoder architecture yields highly effective results across various tasks.
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.