Towards the Generalization of Contrastive Self-Supervised Learning
This work addresses a theoretical gap in self-supervised learning for researchers, offering insights into generalization factors, but it is incremental as it builds on existing contrastive learning methods.
The paper tackles the limited theoretical understanding of contrastive self-supervised learning's generalization ability by defining a measure to quantify data augmentation and providing an upper bound for downstream classification error, revealing it depends on alignment of positive samples, divergence of class centers, and concentration of augmented data, with experiments showing a strong correlation between downstream performance and data concentration.
Recently, self-supervised learning has attracted great attention, since it only requires unlabeled data for model training. Contrastive learning is one popular method for self-supervised learning and has achieved promising empirical performance. However, the theoretical understanding of its generalization ability is still limited. To this end, we define a kind of $(σ,δ)$-measure to mathematically quantify the data augmentation, and then provide an upper bound of the downstream classification error rate based on the measure. It reveals that the generalization ability of contrastive self-supervised learning is related to three key factors: alignment of positive samples, divergence of class centers, and concentration of augmented data. The first two factors are properties of learned representations, while the third one is determined by pre-defined data augmentation. We further investigate two canonical contrastive losses, InfoNCE and cross-correlation, to show how they provably achieve the first two factors. Moreover, we conduct experiments to study the third factor, and observe a strong correlation between downstream performance and the concentration of augmented data.