Responsible AI
summary
Shows concrete tools and results: Azure Machine Learning integrates InterpretML dashboards for per-cohort explainability and Fairlearn for mitigation. A census demo reports overall accuracy 84.4% while initial loan-offer rates differ (males ~23% vs females ~7.36%), and mitigation produces models that trade small accuracy loss for much lower disparity; an EY case drops gender disparity from 7% to 0.5%. Privacy options include the WhiteNoise differential privacy toolkit and upcoming confidential machine learning, plus MLOps audit trails and data sheets.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:18Tech Failures
- 4:41Apple Pay
- 13:09Scandinavian Airlines
- 14:45EY
- 15:59Demo
- 16:14Data
- 26:12Protect
- 28:21Simple Reconstruction
- 38:30Control
- 40:08Failures
- 42:08Machine Learning Models
- 44:37What Can You Do
