Where Did Your Model Learn That? Label-free Influence for Self-supervised Learning
This work addresses the challenge of data attribution in SSL, which is crucial for researchers and practitioners analyzing model behavior, though it is incremental as it adapts existing influence function concepts to the SSL setting.
The authors tackled the problem of understanding how pretraining data influences self-supervised learning (SSL) models by introducing Influence-SSL, a label-free method that identifies influential training examples using representation stability against augmentations, revealing differences in data influence compared to supervised models.
Self-supervised learning (SSL) has revolutionized learning from large-scale unlabeled datasets, yet the intrinsic relationship between pretraining data and the learned representations remains poorly understood. Traditional supervised learning benefits from gradient-based data attribution tools like influence functions that measure the contribution of an individual data point to model predictions. However, existing definitions of influence rely on labels, making them unsuitable for SSL settings. We address this gap by introducing Influence-SSL, a novel and label-free approach for defining influence functions tailored to SSL. Our method harnesses the stability of learned representations against data augmentations to identify training examples that help explain model predictions. We provide both theoretical foundations and empirical evidence to show the utility of Influence-SSL in analyzing pre-trained SSL models. Our analysis reveals notable differences in how SSL models respond to influential data compared to supervised models. Finally, we validate the effectiveness of Influence-SSL through applications in duplicate detection, outlier identification and fairness analysis. Code is available at: \url{https://github.com/cryptonymous9/Influence-SSL}.