Interpretability and explainability in machine learning
summary
Focuses on using SHAP (game‑theory Shapley values) to produce local and global attributions, explains the kernel approximation that samples about 2[M]+2048 coalitions to avoid exponential cost, and shows a live demo with an XGBoost model on the German Credit dataset. Describes packaging and optimizing SHAP in a container, parallelizing jobs with Spark, and the Amazon SageMaker Clarify features: bias checks, baseline choices, explainability outputs, model monitoring charts and reports for right‑to‑explanation compliance.
AI-written summary · about
chapters (8)
- 0:00Introduction to XAI
- 1:30Importance of explainability
- 4:29Methods and categorization
- 7:52Introduction to SHAP
- 9:11Shapley values theory
- 12:44SHAP in machine learning
- 18:11SageMaker Clarify overview
- 22:14Demo and conclusion
related talks
31mBeyond the Black Box: Interpreting ML Models with SHAP
44mImprove explainability of ML models to meet regulatory requirements
35mTaking a peek under the hood - Interpreting black box model
39mExplainable AI as a Conventional Data Analysis Tool
44mInterpretable Machine Learning with Python Examples
35mSHAPtivating Insights: unravelling blackbox AI models
41mMachine Learning Model Fairness and Interpretability with AWS
01h 36mOpen the Black Box: an Introduction to Model Interpretability with LIME and SHAP