Keynote 3: How did you know? Explaining Black Box Model Predictions in Python
summary
Shows concrete tools and outputs: LIME, ELI5, Skater and SHAP produce local explanations, permutation feature importances, partial dependence plots and surrogate decision trees. Cites surprising trust numbers (41.5% can't name a trusted AI app, 9% trust AI with finances, 4% trust AI in hiring) and a working stock-sentiment classifier that needed a language tweak for the word "bomba." Demonstrates credit‑scoring use cases where explanations satisfy regulators and expose bias.
AI-written summary · about
chapters (6)
- 0:00Start
- 3:08What Is Data Science
- 19:14Local Interpretation
- 25:32Permutation Based Feature Importance
- 27:12Partial Dependence Plots
- 30:38Sentiment Analysis for Stocks
related talks
16mBuilding Trust through Explainable AI
46mKeynote: How Can We Trust Machine Learning?
01h 36mOpen the Black Box: an Introduction to Model Interpretability with LIME and SHAP
47mML: Explainable Artificial Intelligence explained
46mAI for High-Stakes Decision Making
33mExplainable AI in the LIME-light
46mCan you trust your AI?
54mKeynote: Can you trust your (large language) model?