Individual Statistical Fairness
summary
Shows that minimizing average classification error can systematically exclude a minority group even when that group is more qualified, and proposes a formal remedy: assume a distribution over tasks and output a mapping from problems to classifiers that equalize individual error rates (probability an individual is misclassified across tasks). Presents an ERM/optimization formulation (agnostic learning setting), proves solvability given a standard learner, and gives generalization bounds that hold when the sample sizes (people and problems) are polynomially large.
AI-written summary · about
chapters (6)
related talks
46mEliciting and Enforcing Subjective Individual Fairness
53mGroups and Individuals in Fairness and Probability
42mFairness in Machine Learning: From Theory to Practice
14mFair And Reliable Machine Learning For High-Stakes Applications:approaches Using Information Theory
43mAfternoon Session: Individuals, Groups, Pattern Recognition and Rich Data
01h 28mMulti-Distribution Learning, for Robustness, Fairness, and Collaboration
46mThe Measure and Mismeasure of Fairness
28mMeasuring Model Fairness