Augmentations vs Algorithms: What Works in Self-Supervised Learning

TL;DR

This paper investigates the relative importance of data augmentations, algorithms, and architectures in self-supervised learning, finding that augmentations and scale are more impactful than algorithmic tweaks.

Contribution

It introduces a unifying framework for SSL methods, enabling direct comparison and revealing that augmentations and scale outperform algorithmic innovations.

Findings

Augmentation techniques significantly improve performance (2-4%).

Algorithmic changes like new losses have minor effects (<1%).

Scaling data and models is more critical than algorithmic improvements.

Abstract

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.