An Engineering Systems Approach to Inference When n Equals One
Abstract
This project will apply techniques from engineering systems to design a novel framework for individualized optimization. This work will focus on individualization over personalization. The former concerns data generated by a single subject, whereas the latter is about taking data from multiple subjects and applying them to an individual. This work consists of three main thrusts. In the first thrust, we will apply compressive sensing to provide new methods for within-subject experiments, known as N-of-1 designs. The goal will be to offer new experiment design plans and new statistical analyses to quickly determine the effectiveness of various treatments. In the second thrust, we will apply online learning and optimal control to study how to find intervention plans to maximize desired outcomes over time. The goal will be to provide protocols that work for a broad class of systems with little identification of the particular dynamics. In the third thrust, we will apply machine learning and reinforcement learning to understand how to learn personalized treatment plans. The goal will be to design within-subject experiments for decision systems, combining past observations of temporal treatment schedules of multiple individuals to hone in on the salient features needed to treat the following subject. The work proposed will draw on mathematical tools, including pattern recognition from machine learning, dynamical system identification from control theory, model-free policy formulation from reinforcement learning, and core mathematics from optimization andsignal processing. If successful, this project could enhance training for sailors, treatment for the injured, and education plans for scholars post-service. Additionally, the work will potentially impact the field of learning robotics and automation, providing new techniques for within-robot calibration and skill acquisition.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 08, 2024
- Source ID
- N000142412531
Entities
People
- Benjamin Recht
Organizations
- Office of Naval Research
- United States Navy
- University of California Regents