Summary of Efficient Bias Mitigation Without Privileged Information, by Mateo Espinosa Zarlenga et al.
Efficient Bias Mitigation Without Privileged Information
by Mateo Espinosa Zarlenga, Swami Sankaranarayanan, Jerone T. A. Andrews, Zohreh Shams, Mateja Jamnik, Alice Xiang
First submitted to arxiv on: 26 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 Deep learning models often exhibit significant performance disparities across groups when trained via empirical risk minimization. This issue arises when group and task labels are spuriously correlated, such as in datasets with grassy backgrounds and cows. Existing bias mitigation methods rely on group labels for training or validation, requiring extensive hyperparameter searches. These limitations hinder practical deployment of these methods, especially when dealing with large datasets, limited computational resources, and complex pipelines. In this paper, we propose Targeted Augmentations for Bias Mitigation (TAB), a simple framework that leverages the entire training history of a helper model to identify spurious samples and generate a group-balanced training set. TAB improves worst-group performance without any group information or model selection, outperforming existing methods while maintaining overall accuracy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making sure deep learning models work fairly for everyone, even when the data has some mistakes. Sometimes, these mistakes can make the model worse at recognizing things that belong to certain groups. The researchers want to fix this problem without needing too much extra information or computation. They created a new way called Targeted Augmentations for Bias Mitigation (TAB) that uses a helper model to find and fix these problems. It makes sure the model is good at recognizing things from all groups, not just one group. |
Keywords
» Artificial intelligence » Deep learning » Hyperparameter
