Bias Amplification Improves Worst-Group Accuracy without Group Information

Gaotang Li, Jiarui Liu, Wei Hu

Keywords: spurious correlation, worst-group accuracy, group robustness

Abstract: Neural networks produced by standard training are known to suffer from poor accuracy on rare subgroups despite achieving high accuracy on average, due to the correlations between certain spurious features and labels. Previous approaches based on worst-group loss minimization (\textit{e.g.} Group-DRO) are effective in improving worse-group accuracy but require expensive group annotations for all the training samples. In this paper, we focus on the more challenging and realistic setting where group annotations are only available on a small validation set or are not available at all. We propose \bam, a novel two-stage training algorithm: in the first stage, the model is trained using a \emph{bias amplification} scheme via introducing a learnable \emph{auxiliary variable} for each training sample together with the adoption of squared loss; in the second stage, we upweight the samples that the bias-amplified model misclassifies, and then continue training the same model on the reweighted dataset. Empirically, \bam leads to consistent improvement over its counterparts in worst-group accuracy, resulting in state-of-the-art performance in spurious correlation benchmarks in computer vision and natural language processing. Moreover, we find a simple stopping criterion that completely removes the need for group annotations, with little or no loss in worst-group accuracy.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning

TL;DR: We propose a novel two-stage training algorithm that achieves the state-of-the-art worst-group accuracy on test data without group information.