Summary of Dataset Representativeness and Downstream Task Fairness, by Victor Borza et al.
Dataset Representativeness and Downstream Task Fairness
by Victor Borza, Andrew Estornell, Chien-Ju Ho, Bradley Malin, Yevgeniy Vorobeychik
First submitted to arxiv on: 28 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computers and Society (cs.CY)
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 This paper explores the connection between dataset representativeness and group-fairness in classifiers trained on those datasets. It highlights a tension between ensuring representative datasets, which can accurately reflect demographic distributions, and training fair classifiers that don’t exhibit biases towards certain groups. The authors demonstrate empirically that better-representative datasets often result in less-fair classifiers, while over-sampling underrepresented groups can lead to greater bias towards those groups. They also investigate fairness-aware sampling strategies, finding that these methods can actually exacerbate the problem by over-sampling majority groups. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how we collect data and train models on it. We often sample people to get a good mix of demographics, but this can lead to biased datasets. This is bad because it can make decisions made by AI algorithms unfair to certain groups. The authors tested different approaches and found that making sure the dataset is representative doesn’t always mean the model will be fair. In fact, some methods even make things worse! They’re trying to figure out how to balance these two important goals so we can have both fairness and representativeness. |
