Fundamental Limits of the Action-Perception Loop
Abstract
In this report, we review progress towards a framework for understanding the foundations of machine learning systems that can interact with their physical environment. We will discuss new results on robust system identification algorithms and methods for safely exploring new policies in uncertain environments. We will describe new progress on collaboration between learning modalities, highlighting techniques for adapting learning systems to new, unseen domains. And we will provide new understanding of machine learning systems that operate on nonconvex losses, showing that these systems necessarily require more computational resources than their linear counterparts.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 11, 2018
- Source ID
- W911NF1610552
Entities
People
- Benjamin Recht
Organizations
- Army Contracting Command
- Defense Advanced Research Projects Agency
- University of California, Berkeley