THE INTEGRATION OF REASONING AND LEARNING STRATEGIES FOR SCIENTIFIC DISCOVERY
Abstract
The integration of learning and reasoning methods can give rise to the next level of AI system performance. The PI obtained preliminary results on AI planning that shows that incorporating deep reinforcement learning into AI planning allows us to solve a class of planning problems that is several orders of magnitude harder than was solvable with current AI planning methods. This research proposes to extend this work to tackle challenge problems in mathematical and scientific discovery. As part of this work, the PI will also develop a curriculum driven machine learning and reasoning framework. Such a framework will enable a system to gradually boost its own reasoning and learning capabilities, starting from novice level performance all the way to super-human level performance.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2021
- Source ID
- FA95502010421
Entities
People
- Bart Selman
Organizations
- Air Force Office of Scientific Research
- Cornell University
- United States Air Force