Explaining model explainability
summary
Explains practical explainability: uses a support vector classifier on the MIMIC MRSA dataset (demo uses 108 samples, ~90% accuracy) and runs explainers from alibi — SHAP (global and local), LIME, Anchors, CEM and counterfactuals — noting SHAP is slow and local versions are cheaper. Shows a concrete surprise: ethnicity and gender rank among top predictors, revealing unintended bias despite high accuracy, and stresses that explainability exposes but does not fix biased models.
AI-written summary · about
chapters (10)
- 0:00Introduction to explainability
- 5:07When explainability is needed
- 9:57Black box models explained
- 12:23When not to use explainability
- 14:28Healthcare use case demo
- 21:51Local explainability algorithms
- 25:41Contrastive explanation methods
- 29:15Live demo with SHAP
- 33:28Recap and key takeaways
- 35:38Q&A on evaluation
related talks
01h 29mNavigating ML Explainability: From Theory to Implementation
24mWhy and How to Build Explainability into your ML Workflow
16mBuilding Trust through Explainable AI
35mWhy Do You Need an Explainable Ai?
46mAI for High-Stakes Decision Making
45mExplainable Machine Learning Explained
39mGuide towards algorithm explainability in machine learning
17mChallenges Towards Production-Ready Explainable Machine Learning