Augmentations vs Algorithms: What Works in Self-Supervised Learning
This work challenges the assumption that algorithmic innovations drive SSL progress, suggesting instead that augmentation diversity and scaling are more critical, which is significant for researchers focusing on efficient SSL development.
The study investigated the relative importance of data augmentations, pretraining algorithms, and model architectures in Self-Supervised Learning, finding that algorithmic improvements often yield minor performance gains (less than 1%), while enhanced augmentations provide more significant improvements (2-4%).
We study the relative effects of data augmentations, pretraining algorithms, and model architectures in Self-Supervised Learning (SSL). While the recent literature in this space leaves the impression that the pretraining algorithm is of critical importance to performance, understanding its effect is complicated by the difficulty in making objective and direct comparisons between methods. We propose a new framework which unifies many seemingly disparate SSL methods into a single shared template. Using this framework, we identify aspects in which methods differ and observe that in addition to changing the pretraining algorithm, many works also use new data augmentations or more powerful model architectures. We compare several popular SSL methods using our framework and find that many algorithmic additions, such as prediction networks or new losses, have a minor impact on downstream task performance (often less than $1\%$), while enhanced augmentation techniques offer more significant performance improvements ($2-4\%$). Our findings challenge the premise that SSL is being driven primarily by algorithmic improvements, and suggest instead a bitter lesson for SSL: that augmentation diversity and data / model scale are more critical contributors to recent advances in self-supervised learning.