Evaluating Self-Supervised Learning via Risk Decomposition
This work addresses the need for better evaluation methods in SSL for researchers and practitioners, offering a more detailed analysis of model performance, though it is incremental as it builds on existing risk decomposition frameworks.
The authors tackled the problem of evaluating self-supervised learning (SSL) models beyond single metrics by proposing a risk decomposition method, which they applied to analyze 169 SSL vision models on ImageNet to identify error sources and provide design insights.
Self-supervised learning (SSL) pipelines differ in many design choices such as the architecture, augmentations, or pretraining data. Yet SSL is typically evaluated using a single metric: linear probing on ImageNet. This does not provide much insight into why or when a model is better, now how to improve it. To address this, we propose an SSL risk decomposition, which generalizes the classical supervised approximation-estimation decomposition by considering errors arising from the representation learning step. Our decomposition consists of four error components: approximation, representation usability, probe generalization, and encoder generalization. We provide efficient estimators for each component and use them to analyze the effect of 30 design choices on 169 SSL vision models evaluated on ImageNet. Our analysis gives valuable insights for designing and using SSL models. For example, it highlights the main sources of error and shows how to improve SSL in specific settings (full- vs few-shot) by trading off error components. All results and pretrained models are at https://github.com/YannDubs/SSL-Risk-Decomposition.