Summary of Measuring and Mitigating Bias For Tabular Datasets with Multiple Protected Attributes, by Manh Khoi Duong et al.
Measuring and Mitigating Bias for Tabular Datasets with Multiple Protected Attributes
by Manh Khoi Duong, Stefan Conrad
First submitted to arxiv on: 29 May 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 A new paper proposes measures and strategies for mitigating discrimination in tabular datasets that contain multiple protected attributes, such as nationality, age, and sex. The authors introduce new discrimination measures and categorize them alongside existing ones to guide researchers and practitioners in assessing the fairness of datasets. They also apply an existing bias mitigation method, FairDo, which can transform a dataset to reduce any type of discrimination, including intersectional discrimination. Experiments on real-world datasets (Adult, Bank, COMPAS) show that de-biasing is possible, with an average reduction in discrimination of 28%. The transformed datasets do not significantly compromise the performances of tested machine learning models compared to the original datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new study helps fix a problem with AI data. Some datasets contain multiple types of information about people, like their age, sex, and nationality. This makes it harder to make sure the data isn’t biased against certain groups. The researchers propose some new ways to measure bias and a way to fix the problem using a technique called FairDo. They tested this on real datasets and found that it works well, reducing bias by an average of 28%. They also found that fixing the bias doesn’t hurt the performance of machine learning models. |
Keywords
» Artificial intelligence » Machine learning
