Removing Unfair Bias in the Machine Learning Pipeline
summary
Demonstrates a concrete pipeline: uses AI Fairness 360 (open source Python) with over 77 fairness metrics and 10 mitigation algorithms to detect and reduce bias, showing COMPAS mislabels African-American defendants at nearly twice the rate of whites and facial‑recognition accuracy 99% for light‑skinned males vs 65% for dark‑skinned females. Shows preprocessing like reweighing driving statistical parity near zero at a small accuracy cost, contrasts pre/in/post processing tradeoffs, and surfaces integration into Watson OpenScale for production monitoring.
AI-written summary · about
chapters (15)
- 0:00Start
- 0:03Trisha Mahoney on Removing Unfair Bias in the Machine Learning Pipeline
- 1:00Intro to Fairness and Bias
- 1:31Why Do We Care about Fairness
- 2:13What Is Fairness
- 5:37Ai Fairness 360 ...
- 6:26Where Bias Comes from and How You Measure
- 6:55Selection Bias
- 8:29Protected Attribute
- 8:55Group Fairness and Individual Fairness
- 9:51Statistical Parity Difference
- 10:21Equal Opportunity Difference
- 15:45Preventing Bias Is Hard
- 17:05Api Documentation
- 21:27German Credit Risk Model
related talks
40mEthical AI: Addressing Bias & Algorithmic Fairness
46mA look at the methods to detect and try to remove bias in ML models
51mDigital Discrimination: Cognitive Bias in Machine Learning
38mSocial Implications of Bias in Machine Learning
47mEthical Machine Learning
01h 30mFairness Tales: How To Measure And Mitigate Unfair Bias in Machine Learning Models
01h 31mHow To Measure And Mitigate Unfair Bias in Machine Learning Models
42mFairness in Machine Learning: From Theory to Practice