Adversarial Machine Learning and Instrumental Variables for Flexible Causal Modeling
summary
Proposes an adversarial minimax objective that converts nonparametric instrumental-variable conditions into a loss minimization problem, proves fast, adaptive rates for the projected RMSE (critical‑radius bounds that recover 1/√n in parametric regimes) by adding a stabilizer = L2 penalty on test functions plus hypothesis regularizers, and shows computational paths: isotonic‑regression oracle for 1D shape constraints, kernel‑trick closed forms for RKHS, and a practical ensemble via FTL+best‑response using regression/classification oracles (random‑forest style) that performs well empirically; warns mean‑square recovery still depends on ill‑posedness (τ_H).
AI-written summary · about
chapters (20)
- 0:00Intro
- 1:51Estimating Price Elasticity of Demand
- 4:14Unobserved Confounding: the major hurdle of causal analysis
- 5:40Instrumental Variables and 2SLS
- 10:14Identification of Causal Effects via Instruments
- 16:38Conditional Moment Restrictions
- 17:29Flexible Causal Models
- 18:41Non-Parametric Instrumental Variable Regression (NPIV)
- 23:41Curse of Dimensionality and Adaptive Estimation
- 24:47Machine Learning to the Rescue
- 25:15Broader Machine Learning Framework
- 26:14Adversarial Estimation of Moment Models
- 27:43Inspired by GANS
- 28:51Main Questions
- 29:36A Rapidly Growing Literature
- 30:12Quality Metric
- 35:06Main Theorem
- 43:19Shape Constraints
- 47:41Synthetic Experiments
- 49:04Reproducing Kernel Hilbert Spaces
related talks
44mInvariance, Causality and Novel Robustness
01h 11mA Theory of Multi-objective Machine Learning
43mRobustness in High-Dimensional Inference Tasks
09mTowards Credible and Effective Data-Driven Decision-Making
01h 21mAdversarial Bandits: Theory and Algorithms
30mKernel Methods For Causal Inference
01h 23mImplicit Regularization II
12mCausal Inference And Data-Driven Decision-Making