Eliciting and Enforcing Subjective Individual Fairness
summary
Proposes an algorithm that enforces pairwise subjective similarity constraints without assuming any metric, trading off classification error against the frequency of expert complaints; implements an Oracle-efficient reduction via Lagrangian duality and multiplicative-weights best-response iterations to solve weighted classification subproblems; proves VC-dimension based generalization and that only linearly many pairwise queries suffice; runs preliminary experiments on the COMPAS dataset with 43 undergraduates as judges and shows empirical convergence (example: gamma = 30%, ADA = 0).
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:12Definitions
- 1:50Family of Statistical Fairness Measures
- 6:50Fairness Through Awareness Paper
- 19:00The Cons
- 19:33The Approach
- 20:56Related Work
related talks
53mGroups and Individuals in Fairness and Probability
43mMetric Learning for Individual Fairness
30mIndividual Statistical Fairness
54mKeynote - Fairness and Diversity in Online Social Systems
43mAfternoon Session: Individuals, Groups, Pattern Recognition and Rich Data
34mInterpretability and Algorithmic Fairness
48mPersonalized Collaborative Learning with Affinity-Based Variance Reduction
14mFair And Reliable Machine Learning For High-Stakes Applications:approaches Using Information Theory