Machine Intelligence
Abstract
The objective of this effort was to produce technologies that could reduce manpower requirements and improve response times of future command and control (C2) systems through research in machine learning. This in-house research effort explored various areas of machine learning to produce technologies capable of supporting future C2 systems. Machine learning has the potential to reduce manpower requirements, reduce decision cycle times, and improve the robustness of C2 systems. However, many obstacles, such as intractability in large state spaces, prevent the application of these technologies to practical C2 problems. The goal, then, was to research and develop innovative technologies that overcome said obstacles and enable the application of machine learning to relevant C2 problems. This goal was achieved through the research and development of new state space abstraction and feature selection algorithms. Theoretical and empirical results of this effort were published to refereed conferences and showcased in technology demonstrations. In this document, we detail our approaches and report our results in improved scaling of reinforcement learning via feature set reduction and state space abstraction.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2013
- Accession Number
- ADA580353
Entities
People
- Nathaniel Gmelli
- Robert Wright
- Steven Loscalzo
Organizations
- Air Force Research Laboratory