The SSL Interplay: Augmentations, Inductive Bias, and Generalization
This work addresses practical challenges in SSL for engineers, but it appears incremental as it builds on existing theory without introducing a new method.
The paper tackles the instability and representation collapse issues in self-supervised learning by analyzing how data augmentation, network architecture, and training algorithms affect generalization, providing insights for practitioners.
Self-supervised learning (SSL) has emerged as a powerful framework to learn representations from raw data without supervision. Yet in practice, engineers face issues such as instability in tuning optimizers and collapse of representations during training. Such challenges motivate the need for a theory to shed light on the complex interplay between the choice of data augmentation, network architecture, and training algorithm. We study such an interplay with a precise analysis of generalization performance on both pretraining and downstream tasks in a theory friendly setup, and highlight several insights for SSL practitioners that arise from our theory.