Lightweight representations for decentralized learning in data-rich environments

Abstract

Agents operating autonomously -- as in the cyber or physical domains of interest to the Navy --can be expected to collect massive amounts of heterogeneous data. In theory, this richness of data should be a boon to learning, reasoning, and dealing with the uncertainty of these environments, as well as fruitful communication between agents. In practice, the sheer scale of data presents a number of challenges. For instance, (1) storing all of the data in memory locally may be prohibitive, especially for small sensors or drones; (2) the computing power and energy required to process and apply modern, complex machine learning algorithms to this data may be prohibitive; (3) the bandwidth required to communicate this data wholesale to other agentsmay be prohibitive -- especially in contested or uncertain environments. To address these concerns, we propose to construct lightweight representations of the collected data instead ofworking with the full data set. Our proposed summaries are lightweight from all three of the memory, learning, and bandwidth perspectives. They enable efficient and reliable data processing and machine learning -- as well as efficient and reliable self-monitoring mechanisms. To construct our summaries, we use ideas from machine learning, statistics, theoretical computer science, computational geometry, statistical physics, perturbation theory, and network analysis. Approved for Public Release

Document Details

Document Type
DoD Grant Award
Publication Date
Jun 17, 2020
Source ID
N000142012532

Entities

People

  • Tamara Broderick

Organizations

  • Massachusetts Institute of Technology
  • Office of Naval Research
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Distributed Systems and Data Platform Development
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • Autonomy
  • Cyber