Collaborative Proposal: Contested Logistics
Abstract
We will develop a suite of high-performance analytics tools to assist U.S. Navy planners in the rapid development of operational and logistical models that are resilient toward adversarial action and are capable of live reaction to real-time developments. Our tools will (a) enable the rapid evaluation of a proposed Blue plan against a Red adversary whose capabilities and goals are not precisely known, so as to expose vulnerabilities of said plan as a function of intensity, location and time, and (b) develop Red-resilient Blue plans that incorporate recourse, i.e., adapt as a function of prior and ongoing Red actions, both observed and estimated. An essential goal of our proposed work is that we seek to deliver tools that, post-development, run quickly even on a modest computing platform. We will also deliver versions of our tools, to be used for strategic planning, that are suited for massive computing and simulation, including deep learning, reinforcement learning, and related methodologies. We view as crucially important that our tools enable (human) decision-makers without optimization or machine-learning backgrounds to actually be able to critically evaluate scenarios discovered by our algorithms.Our methodology will encompass ideas derived from adroit use of the robust optimization paradigm --which we have deployed in prior work-- adapted to high-performance computing (in particular, GPU-based computing), and ideas from modern developments in supervised reinforcement learning, combined with deep learning, as has been applied in recent advancesin game learning without human input.Approved for public release.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 08, 2024
- Source ID
- N000142412443
Entities
People
- Daniel Bienstock
Organizations
- Office of Naval Research
- Trustees of Columbia University in the City of New York
- United States Navy