Summary of Mitigating Spurious Correlations Via Disagreement Probability, by Hyeonggeun Han et al.
Mitigating Spurious Correlations via Disagreement Probability
by Hyeonggeun Han, Sehwan Kim, Hyungjun Joo, Sangwoo Hong, Jungwoo Lee
First submitted to arxiv on: 4 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper addresses the issue of biased models trained using empirical risk minimization (ERM) being prone to poor performance when there are no spurious correlations between target labels and bias attributes. A novel training objective is introduced, which aims to robustly enhance model performance across all data samples, regardless of the presence of spurious correlations. This leads to a debiasing method called Disagreement Probability based Resampling for debiasing (DPR), which leverages the disagreement between target labels and biased model predictions to identify bias-conflicting samples and upsamples them according to the disagreement probability. The paper evaluates DPR on multiple benchmarks, demonstrating state-of-the-art performance over existing baselines that do not use bias labels. Additionally, a theoretical analysis is provided, detailing how DPR reduces dependency on spurious correlations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper finds a way to make machine learning models fairer by fixing a problem with how they’re trained. Currently, these models can be biased towards certain things just because those things are related to the target we want them to learn. This is bad news if we don’t have labels for what’s causing that bias. The solution is called Disagreement Probability based Resampling for debiasing (DPR). It looks at when a model makes mistakes and uses that information to make sure it learns from the right examples. Tests show that this method works really well and can even do better than other methods that require special labels. |
Keywords
» Artificial intelligence » Machine learning » Probability
