Learning Techniques for Planning

Tutorial given at ICAPS 2007

Sungwook Yoon & Subbarao Kambhampati
School of Computing & Informatics
Arizona State University

Abstract:

This tutorial aims to provide an overview of learning techniques for planning--an area with significant past as well as a substantial current interest. From planning perspective, learning techniques can be seen as (i) a way to "speed-up" planners or (ii) as a way to reduce domain-modeling burden. Early interest in this area was driven almost exclusively by the first, while the recent resurgence is motivated by the second consideration. From the learning perspective, focus on planning necessitates a welcome diversion from the dominant tabula-rasa paradigms. Specifically, it encourages handling relational and first order representations, and foregrounds the need for knowledge-intensive learning techniques (such as explanation-based and relevance-based learning).

We will survey both the history as well as promising current trends, and make connections to DARPA Integrated Learning program, Model-lite planning.

Presentation Material (from ICAPS 2007)

Resources

A set of lectures on learning techniques in planning delivered by Subbarao Kambhampati at 2006 Machine Learning Summer School.
A recent survey article we wrote for the AI Magazine titled "Learning-Assisted Automated Planning" ( a copy of which is available here ).
Model-Lite Planning: Resources
Here is an earlier tutorial (by Veloso & Borrajo) on learning methods in planning.
Here is a resource list for papers and researchers in Learning for planning.

Subbarao Kambhampati

Last modified: Thu Sep 27 12:44:35 MST 2007