Outlier-Aware Training for Improving Group Accuracy Disparities
Not on the map. Why?
Abstract (arXiv)
Methods addressing spurious correlations such as Just Train Twice (JTT, arXiv:2107.09044v2) involve reweighting a subset of the training set to maximize the worst-group accuracy. However, the reweighted set of examples may potentially contain unlearnable examples that hamper the model's learning. We propose mitigating this by detecting outliers to the training set and removing them before reweighting. Our experiments show that our method achieves competitive or better accuracy compared with JTT and can detect and remove annotation errors in the subset being reweighted in JTT.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 12 of its 19 references that are in the map. First posted on arXiv 27 Oct 2022.
Builds on
12 of 19 references.
References · 12
- 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
- 2017A Broad-Coverage Challenge Corpus for Sentence Understanding through Inferencecited by 2,682
- 2018A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attackscited by 1,485
- 2018FEVER: a large-scale dataset for Fact Extraction and VERificationcited by 1,375
- 2018Annotation Artifacts in Natural Language Inference Datacited by 671
- 2021Just Train Twice: Improving Group Robustness without Training Group Informationcited by 523
- 2020Gradient Starvation: A Learning Proclivity in Neural Networkscited by 267
- 2021Get Your Vitamin C! Robust Fact Verification with Contrastive Evidencecited by 225
- 2020Large-Scale Methods for Distributionally Robust Optimizationcited by 179
- 2021Simple data balancing achieves competitive worst-group-accuracycited by 178
- 2019Towards Debiasing Fact Verification Modelscited by 143
- 2021Predict then Interpolate: A Simple Algorithm to Learn Stable Classifierscited by 16
Built on by
Not cited yet.