Ideas on Machine Learning Interpretability
5 likes833 views
The document discusses various approaches to machine learning interpretability, emphasizing the importance of transparency, trust, and fairness in model predictions. It explores the complexities of learned functions, model-specific and model-agnostic techniques, and the challenges associated with linear and nonlinear models. Key concepts include global vs. local interpretability, variable importance measures, and methods for providing explanations to enhance user understanding and decision-making.
Ideas on Machine Learning Interpretability
- 2. Big Ideas
- 3. Learning from data … Adapted from: Learning from Data. https://work.caltech.edu/textbook.html
- 4. EXPLAIN HYPOTHESIS h ≈ g, βj g(x(i) j), g(x(i) (-j)) (explain predictions with reason codes) Learning from data … transparently. Adapted from: Learning from Data. https://work.caltech.edu/textbook.html
- 7. 7 A framework for interpretability Complexity of learned functions: • Linear, monotonic • Nonlinear, monotonic • Nonlinear, non-monotonic (~ Number of parameters/VC dimension) Enhancing trust and understanding: the mechanisms and results of an interpretable model should be both transparent AND dependable. Understanding ~ transparency Trust ~ fairness and accountability Scope of interpretability: Global vs. local Application domain: Model-agnostic vs. model-specific
- 10. Age Numberof Purchases Lost profits. Wasted marketing. “For a one unit increase in age, the number of purchases increases by 0.8 on average.” 𝑔 𝑥 = 0.8 𝑥 Linear Models Machine Learning Exact explanations for approximate models. Approximate explanations for exact models. Age “Slope begins to decrease here. Act to optimize savings.” “Slope begins to increase here sharply. Act to optimize profits.” Numberof Purchase 𝑔 𝑥 ≈ 𝑓(𝑥)
- 12. Partial dependence plots Source: http://statweb.stanford.edu/~tibs/ElemStatLearn/printings/ESLII_print10.pdf HomeValue ~ MedInc + AveOccup + HouseAge + AveRooms
- 13. Surrogate models
- 21. Resources
- 23. Questions?