Using Qualitative Models to Localize Reinforcement Learning

Problem
Reinforcement learning is a powerful, numerical, data-intensive and general-purpose learning methods. It is also computationally very costly.  One possible was of addressing this cose might be to use knowledge-intensive, symbolic models to partition the search space and run reinforcement learning only in small search spaces.

Students will first read relevant papers and then design a scenario in which symbolic reasoning may guide numerical learning.