High-Dimensional Causal Prediction
Abstract
We propose to research algorithms for estimation of individual treatment effect from highdimensional observational data.1. The cont"ractor shall conduct research on learning representations for counterfactual inference, studyinginterpretability, uncertainty, mult""iple structured and/or continuous treatments, sample complexity, and robustness to hidden confounding.2. The contractor shall condu""ct research on hidden confounding with high-dimensional proxies, providing theoretical conditions of identifiability in the presence" of hidden confounders and practical algorithms based on variational autoencoders.3. The contractor shall conduct research on disco"vering treatment regions in settings where there is substantially more control than treated data, giving efficient algorithms that c"an identify responders and the control and treatment response functions.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 01, 2017
- Source ID
- N000141712791
Entities
People
- David A Sontag
Organizations
- Massachusetts Institute of Technology
- Office of Naval Research
- United States Navy