Stochastic Contrastive Learning
This work addresses the need for more interpretable and compressed representations in self-supervised learning, offering incremental improvements for tasks like classification and regression.
The paper tackled the problem of contrastive self-supervised learning models lacking latent variable inference by introducing latent variable approximations, resulting in improved downstream performance (e.g., 96.42% on CIFAR10 and 77.49% on ImageNet) and highly compressed representations (588x reduction).
While state-of-the-art contrastive Self-Supervised Learning (SSL) models produce results competitive with their supervised counterparts, they lack the ability to infer latent variables. In contrast, prescribed latent variable (LV) models enable attributing uncertainty, inducing task specific compression, and in general allow for more interpretable representations. In this work, we introduce LV approximations to large scale contrastive SSL models. We demonstrate that this addition improves downstream performance (resulting in 96.42% and 77.49% test top-1 fine-tuned performance on CIFAR10 and ImageNet respectively with a ResNet50) as well as producing highly compressed representations (588x reduction) that are useful for interpretability, classification and regression downstream tasks.