The Elephant in your Dataset: Addressing Bias in Machine Learning
summary
Defines seven injection points for harmful bias (historical, representation, measurement, aggregation, learning, evaluation, deployment) citing a 2021 paper, runs a logistic regression on the UCI Adult dataset with ~84% accuracy, uses Microsoft’s Responsible AI Toolkit fairness dashboard to compare metrics, demonstrates pre-processing balancing and counterfactual augmentation, shows demographic parity difference drop from 19% to 1.08% and equalized odds difference from 15.3% to 5.16%, warns counterfactual flips labels so is risky for production.
AI-written summary · about
chapters (8)
- 0:00Introduction and agenda
- 2:20Intersectionality exercise
- 4:40Defining bias and terminology
- 7:00Sources of harmful bias
- 18:42Fairness and metrics
- 23:48Mitigation strategies
- 25:54Practical demo and analysis
- 37:23Resources and conclusion
related talks
47mThe Elephant in your Dataset Addressing Bias in Machine Learning
51mDigital Discrimination: Cognitive Bias in Machine Learning
23mRemoving Unfair Bias in the Machine Learning Pipeline
01h 00mThe Elephant in your Dataset: Addressing Bias in Machine Learning
38mSocial Implications of Bias in Machine Learning
50mEthical Machine Learning
47mEthical Machine Learning
42mTowards Inclusive AI: Understanding bias in ML models