Online Program

Deep Doubly Robust Outcome-Weighted Learning (308334)

*Xiaotong Jiang , University of North Carolina at Chapel Hill 
Michael Kosorok, University of North Carolina at Chapel Hill 

Keywords: Machine Learning, Deep Learning, Precision Medicine, Decision Making

We propose a new precision medicine approach called deep doubly robust outcome weighted learning (DDROWL) that applies deep learning (DL) techniques. DDROWL is a new machine-learning tool that estimates the optimal decision rule and achieves the best of three worlds: double robustness, residual weighted learning, and DL. With the implementation of neural networks, DDROWL is able to expand the influence of precision medicine to high-dimensional data with great flexibility and computation power. We are able to confirm that DL models are better at extracting representations from abundant high-dimensional data whereas the classical L1-PLS and RF models are eminent, computationally fast methods for small to medium or large and sparse data.