FroSSL: Frobenius Norm Minimization for Efficient Multiview Self-Supervised Learning

TL;DR

FroSSL introduces a novel Frobenius norm-based objective for multiview self-supervised learning that improves training efficiency and convergence speed without eigendecomposition, achieving competitive results on multiple datasets.

Contribution

FroSSL proposes a new objective function that combines covariance regularization and augmentation invariance efficiently, avoiding eigendecomposition and enhancing training speed in SSL.

Findings

FroSSL converges faster than existing SSL methods.

FroSSL achieves competitive accuracy on multiple datasets.

Theoretical analysis links FroSSL's efficiency to eigenvalue behavior.

Abstract

Self-supervised learning (SSL) is a popular paradigm for representation learning. Recent multiview methods can be classified as sample-contrastive, dimension-contrastive, or asymmetric network-based, with each family having its own approach to avoiding informational collapse. While these families converge to solutions of similar quality, it can be empirically shown that some methods are epoch-inefficient and require longer training to reach a target performance. Two main approaches to improving efficiency are covariance eigenvalue regularization and using more views. However, these two approaches are difficult to combine due to the computational complexity of computing eigenvalues. We present the objective function FroSSL which reconciles both approaches while avoiding eigendecomposition entirely. FroSSL works by minimizing covariance Frobenius norms to avoid collapse and minimizing mean-squared error to achieve augmentation invariance. We show that FroSSL reaches competitive accuracies more quickly than any other SSL method and provide theoretical and empirical support that this faster convergence is due to how FroSSL affects the eigenvalues of the embedding covariance matrices. We also show that FroSSL learns competitive representations on linear probe evaluation when used to train a ResNet-18 on several datasets, including STL-10, Tiny ImageNet, and ImageNet-100.