Interpretability and Algorithmic Fairness
summary
Two short projects: one formalizes causal fairness with a potential outcomes framework that shifts interventions from immutable attributes to the perception of those attributes, maps timing (pre/post interview) to different estimands, and uses a synthetic hiring simulation with parameters alpha/beta/gamma to show causal and statistical parity metrics diverge when baseline qualifications differ. The second trains a triplet-based embedding from human odd-one-out judgments to surface interpretable visual axes (race, age, hair, expression), finds high human–model placement correlation, and documents annotator-dependent biases.
AI-written summary · about
chapters (4)
- 0:00Introduction and welcome
- 0:39Causal fairness in ML
- 19:20Visual diversity auditing
- 30:34Q&A and session wrap-up
related talks
35mOpen Source Fairness
42mFairness in Machine Learning: From Theory to Practice
37mAlgorithmic Fairness From The Lens Of Causality And Information Theory
36mMachine Learning Fairness: Lessons Learned
44mProcedural Justice in Algorithmic Fairness: Transparency, Control and Participation Perspectives
40mQuestions and Practices to Make Algorithmic Decision Making More Fair
37mKeynote: Ethical Machine Learning: Creating Fair Models in an Unjust World
14mFair And Reliable Machine Learning For High-Stakes Applications:approaches Using Information Theory