Causal Inference And Data-Driven Decision-Making
summary
Presents a concrete method for learning treatment rules that is robust to unobserved confounding by optimizing worst‑case policy value over an ambiguity set around estimated propensities, using minimax policy learning with inverse-propensity weighting and L_infty bounds on selection bias. Gives finite‑sample convergence rates and decision‑theoretic regret guarantees, scales the estimator to sequential settings (infinite‑horizon MDPs) under per‑step independent confounding, and shows practical relevance to personalized medicine and dosing.
AI-written summary · about
chapters (5)
related talks
41mKeynote: Personalization, Bandits, and Causal inference
28mConfounding-Robust Policy Evaluation In Infinite-Horizon Reinforcement Learning
39mCausal Effect Estimation in Practice: Lessons Learned from E-commerce & Banking
40mDecision-Making Under Miscalibration
30mNon-parametric Causal Inference in Dynamic Thresholding Designs
41mOn the Path to Causal Inference
53mAdversarial Machine Learning and Instrumental Variables for Flexible Causal Modeling
17mCodifying Data Science Intuition