Bias in Planning and Explanation-Based Learning
Abstract
Biases enable systems to make decisions in realms where all legitimate sources of knowledge have been exhausted. This article investigates the application of biases to the problem of planning, and how this can indirectly induce effective biases in a learning process that is based on planner's experiences. Experimental results from six biased planners, plus several more complex multimethod planners, indicate complex tradeoffs among planner completeness, planning efficiency, and length. Learning also varies in complex ways among these planners, with one notable result being the ease with which some planners learn rules that can generalize from one object to many; a phenomenon known in machine learning as generalization to N. Bias, Planning, Explanation-based learning, Multi-method planners, Soar.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1993
- Accession Number
- ADA269608
Entities
People
- Amy Unruh
- Paul Simon Rosenbloom
- Soowon Lee
Organizations
- University of Southern California