Robust Methods for Trustworthy Machine Learning

Abstract

The overarching goal of this proposed research project is to greatly enhance the robustness of existing machine learning systems so that they are adaptive to resource-limited or time-critical contexts. Specifically, the team will develop novel mathematical theories and computational algorithms for 1) robust Monte Carlo modeling and simulation techniques to empower various optimization tasks, 2) robust multimodal representation so that diverse data collected from a network of autonomous learning units can be reliably integrated, 3) robust anomaly detection to online monitor potential threats as revealed from the underlying data patterns, 4) robust prediction in the context of time-varying data generating processes, 5) systematic evaluation of robustness with theory-guided diagnostic tools. The team will demonstrate their work with data from various domains, including extreme weather prediction, sensor monitoring, and autonomous driving. The various robust learning methods developed under this project will greatly facilitate trustworthy learning in many Army-unique research projects, which can further enable future Army force modernization capabilities.

Document Details

Document Type
DoD Grant Award
Publication Date
Jul 09, 2020
Source ID
W911NF2010222

Entities

People

  • Jie Ding

Organizations

  • Army Contracting Command
  • United States Army
  • University of Minnesota

Tags

Readers

  • Computational Fluid Dynamics (CFD)
  • Distributed Systems and Data Platform Development
  • Military Science and Technology Research and Modernization.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks