Foundations and Applications of Dynamics in Interventional Learning
Abstract
This project will explore the foundations of machine learning systems that can intervene and act on the physical world. Unlike traditional passive learning or active learning, these systems change the state of the system that they interact with, and hence pose new challenges for algorithm design, controller design, and measurement.This project will explore the contours of interventional learning through three closely interacting themes. First, we will explore the notion of sample complexity in the interventional setting. We will formalize the appropriate notions of complexity and power needed to identify dynamical systems for the purpose of controlling state or enhancing measurement. This will require merging tools from statistical learning theory with those from robust control. Second, we will build a framework for analyzing machine learning algorithms as dynamical systems. Using stability analysis tools from controls, we will determine how to automatically verify the convergence of algorithms to desired optimal
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 03, 2017
- Source ID
- N000141712191
Entities
People
- Benjamin Recht
Organizations
- Office of Naval Research
- United States Navy
- University of California Regents