On Feature Decorrelation in Self-Supervised Learning
This addresses a subtle but impactful issue in self-supervised learning for AI researchers, though it is incremental as it builds on existing frameworks.
The paper tackled the problem of dimensional collapse in self-supervised learning, where features become strongly correlated, and found that feature decorrelation improves representation quality, with empirical gains demonstrated.
In self-supervised representation learning, a common idea behind most of the state-of-the-art approaches is to enforce the robustness of the representations to predefined augmentations. A potential issue of this idea is the existence of completely collapsed solutions (i.e., constant features), which are typically avoided implicitly by carefully chosen implementation details. In this work, we study a relatively concise framework containing the most common components from recent approaches. We verify the existence of complete collapse and discover another reachable collapse pattern that is usually overlooked, namely dimensional collapse. We connect dimensional collapse with strong correlations between axes and consider such connection as a strong motivation for feature decorrelation (i.e., standardizing the covariance matrix). The gains from feature decorrelation are verified empirically to highlight the importance and the potential of this insight.