UnLearning from Experience to Avoid Spurious Correlations
On the map: LLM safety and security → Machine unlearning
Abstract (arXiv)
While deep neural networks can achieve state-of-the-art performance in many tasks, these models are more fragile than they appear. They are prone to learning spurious correlations in their training data, leading to surprising failure cases. In this paper, we propose a new approach that addresses the issue of spurious correlations: UnLearning from Experience (ULE). Our method is based on using two classification models trained in parallel: student and teacher models. Both models receive the same batches of training data. The student model is trained with no constraints and pursues the spurious correlations in the data. The teacher model is trained to solve the same classification problem while avoiding the mistakes of the student model. As training is done in parallel, the better the student model learns the spurious correlations, the more robust the teacher model becomes. The teacher model uses the gradient of the student's output with respect to its input to unlearn mistakes made by the student. We show that our method is effective on the Waterbirds, CelebA, Spawrious and UrbanCars datasets.
Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Machine unlearning, one of 12 directions in LLM safety and security. Builds on 40 of its 54 references that are in the map. First posted on arXiv 4 Sep 2024.
Builds on
40 of 54 references.
References · 40
- 2015Deep Residual Learning for Image Recognitioncited by 51,809
- 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
- 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,690
- 2014Deep Learning Face Attributes in the Wildcited by 3,679
- 2019Invariant Risk Minimizationcited by 1,832
- 2019Self-training with Noisy Student improves ImageNet classificationcited by 1,304
- 2016Deep CORAL: Correlation Alignment for Deep Domain Adaptationcited by 1,194
- 2018Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labelscited by 994
- 2020Out-of-Distribution Generalization via Risk Extrapolation (REx)cited by 737
- 2021Just Train Twice: Improving Group Robustness without Training Group Informationcited by 522
- 2015Return of Frustratingly Easy Domain Adaptationcited by 389
- 2022Last Layer Re-Training is Sufficient for Robustness to Spurious Correlationscited by 379
28 more
- 2020The Pitfalls of Simplicity Bias in Neural Networkscited by 349
- 2014CNN Features off-the-shelf: an Astounding Baseline for Recognitioncited by 316
- 2020Environment Inference for Invariant Learningcited by 306
- 2021Gradient Matching for Domain Generalizationcited by 258
- 2019Adversarial Domain Adaptation with Domain Mixupcited by 250
- 2020No Subclass Left Behind: Fine-Grained Robustness in Coarse-Grained Classification Problemscited by 208
- 2022Improving Out-of-Distribution Robustness via Selective Augmentationcited by 202
- 2022Correct-N-Contrast: A Contrastive Approach for Improving Robustness to Spurious Correlationscited by 171
- 2022On Feature Learning in the Presence of Spurious Correlationscited by 145
- 2020Learning from Failure: Training Debiased Classifier from Biased Classifiercited by 133
- 2022Revisiting Weakly Supervised Pre-Training of Visual Perception Modelscited by 116
- 2022Spread Spurious Attribute: Improving Worst-group Accuracy with Spurious Attribute Estimationcited by 88
- 2022A Whac-A-Mole Dilemma: Shortcuts Come in Multiples Where Mitigating One Amplifies Otherscited by 85
- 2022Agree to Disagree: Diversity through Disagreement for Better Transferabilitycited by 69
- 2023Simple and Fast Group Robustness by Automatic Feature Reweightingcited by 68
- 2020The Surprising Simplicity of the Early-Time Learning Dynamics of Neural Networkscited by 58
- 2022ZIN: When and How to Learn Invariance Without Environment Partition?cited by 58
- 2022Diversify and Disambiguate: Learning From Underspecified Datacited by 55
- 2023Identifying Spurious Biases Early in Training through the Lens of Simplicity Biascited by 44
- 2023Spawrious: A Benchmark for Fine Control of Spurious Correlation Biasescited by 41
- 2022Invariant Causal Mechanisms through Distribution Matchingcited by 31
- 2023Overcoming Simplicity Bias in Deep Networks using a Feature Sievecited by 29
- 2022The Two Dimensions of Worst-case Training and the Integrated Effect for Out-of-domain Generalizationcited by 21
- 2022You Only Need a Good Embeddings Extractor to Fix Spurious Correlationscited by 17
- 2024Out of Spuriousity: Improving Robustness to Spurious Correlations without Group Annotationscited by 9
- 2023Even Small Correlation and Diversity Shifts Pose Dataset-Bias Issuescited by 8
- 2023Few-shot Adaptation to Distribution Shifts By Mixing Source and Target Embeddingscited by 3
- 2023Identifying and Disentangling Spurious Features in Pretrained Image Representationscited by 3
Built on by
Citing papers per year: 2025: 1; 2026: 1.
2 papers.