Understanding self-supervised Learning Dynamics without Contrastive Pairs
This work addresses a fundamental theoretical gap in self-supervised learning for researchers, providing insights into non-contrastive methods, though it is incremental as it builds on existing non-contrastive approaches.
The paper tackles the problem of understanding why non-contrastive self-supervised learning methods avoid representational collapse without negative pairs, proposing DirectPred, a novel approach that sets the linear predictor based on input statistics without gradient training. On ImageNet, it performs comparably to complex predictors and outperforms a linear predictor by 2.5% in 300-epoch training and 5% in 60-epoch training.
While contrastive approaches of self-supervised learning (SSL) learn representations by minimizing the distance between two augmented views of the same data point (positive pairs) and maximizing views from different data points (negative pairs), recent \emph{non-contrastive} SSL (e.g., BYOL and SimSiam) show remarkable performance {\it without} negative pairs, with an extra learnable predictor and a stop-gradient operation. A fundamental question arises: why do these methods not collapse into trivial representations? We answer this question via a simple theoretical study and propose a novel approach, DirectPred, that \emph{directly} sets the linear predictor based on the statistics of its inputs, without gradient training. On ImageNet, it performs comparably with more complex two-layer non-linear predictors that employ BatchNorm and outperforms a linear predictor by $2.5\%$ in 300-epoch training (and $5\%$ in 60-epoch). DirectPred is motivated by our theoretical study of the nonlinear learning dynamics of non-contrastive SSL in simple linear networks. Our study yields conceptual insights into how non-contrastive SSL methods learn, how they avoid representational collapse, and how multiple factors, like predictor networks, stop-gradients, exponential moving averages, and weight decay all come into play. Our simple theory recapitulates the results of real-world ablation studies in both STL-10 and ImageNet. Code is released https://github.com/facebookresearch/luckmatters/tree/master/ssl.