FACTS: First Amplify Correlations and Then Slice to Discover Bias
This addresses the issue of model bias due to spurious correlations in datasets, which is crucial for improving fairness and robustness in computer vision, though it is an incremental advance on existing bias identification methods.
The paper tackles the problem of identifying bias-conflicting data slices in computer vision datasets where spurious correlations exist, and proposes FACTS, a method that first amplifies correlations and then slices to discover bias, achieving up to 35% improvement in precision@10 over prior work.
Computer vision datasets frequently contain spurious correlations between task-relevant labels and (easy to learn) latent task-irrelevant attributes (e.g. context). Models trained on such datasets learn "shortcuts" and underperform on bias-conflicting slices of data where the correlation does not hold. In this work, we study the problem of identifying such slices to inform downstream bias mitigation strategies. We propose First Amplify Correlations and Then Slice to Discover Bias (FACTS), wherein we first amplify correlations to fit a simple bias-aligned hypothesis via strongly regularized empirical risk minimization. Next, we perform correlation-aware slicing via mixture modeling in bias-aligned feature space to discover underperforming data slices that capture distinct correlations. Despite its simplicity, our method considerably improves over prior work (by as much as 35% precision@10) in correlation bias identification across a range of diverse evaluation settings. Our code is available at: https://github.com/yvsriram/FACTS.