Action-Space Partitioning for Planning
Natalia H. Gardiol
MIT Computer Science and Artificial Intelligence Lab
February 16, 2007
Talk: 11:00 am - 12:00 pm
Location: 3105 Engineering
Host: Laura K. Dillon
Abstract:
For autonomous artificial decision-makers to solve realistic tasks, they need to deal with searching through large state and action spaces under time pressure. We study the problem of planning in such domains and show how structured representations of action effects can help partition the action space into a set of equivalence classes at run time. The pared-down action space is used to identify an informative subset of the state space in which to search for a solution. This analysis can yield large gains in planning efficiency.
Biography:
Natalia Gardiol is a PhD student in the Computer Science and Artificial Intelligence Laboratory (CSAIL) at the Massachusetts Institute of Technology (MIT). She received a B.S. in Computer Science in 1999 from Michigan State University and an M.S. in 2003 from MIT. She has done research in reinforcement learning, representations for learning, decision-theoretic planning, and will receive her PhD from MIT in 2007.