A Platform for Computational-and-Data-Intensive Methods for Large-Scale Intelligent Distributed Systems
Abstract
We request a compute cluster to support research on computational-and-data intensive methods for large-scale intelligent distributed systems for multi-agent decision making and scientific discovery. Our research involves the synthesis of algorithmic development with empirical validation, and builds on connections between different research areas Ñ in particular, artificial intelligence and machine learning, operations research, game theory, network science, complex adaptive systems, and human computation. Central themes of our research are: (1) Constraint reasoning, optimization, and machine learning for solving hard combinatorial problems; (2) Exploiting problem structure; (3) Human computation for boosting combinatorial solvers; and (4) Citizen science and crowdsourcing for data collection for scientific discovery. As a key application domain, we consider scientific discovery for materials science, an effort currently funded by ARO grant W911NF-14-1-0498. Our approach tightly couples constraint reasoning, probabilistic reasoning, and machine learning techniques combined with human insights. Moreover, our proposed computational framework is general and applicable to a range of domains. We also develop new methods for reasoning under uncertainty in multi-agent adversarial settings, new formalisms for integrating structure obtained from data via deep learning with inference technology, as well new approaches to factor in insights from typical case complexity. With the recent dramatic advances in deep learning, an integration with general reasoning and decision making methods holds the promise of a new level of capabilities for autonomous intelligent systems, which would have significant applications across a range of DoD domains. For example, the requested computational resources will allow us to scale up multi-agent reasoning to thousands of interacting agents, where each agent is modeled with its own set of incentives and objectives, learned from data about previous interactions.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 11, 2018
- Source ID
- W911NF1710187
Entities
People
- Carla Gomes
Organizations
- Army Contracting Command
- Cornell University
- United States Army