Share Email Print

Proceedings Paper

Learning Plans Through Experience: A First Pass In The Chess Domain
Author(s): James C. Spohrer
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

Nearly all successful programs which play chess rely heavily on a brute-force search of thousands or millions of positions in a game tree. Human experts, on the other hand, rely more on plans to guide their search for an appropriate move. In addition, human experts began as novices and had to learn the chess plans they use. Just as novice chess players construct plans as they play the game, an intelligent robot should be able to learn a library of plans for a domain through experience. In this paper we propose a three step process for acquiring new plans through experience, and describe a program which uses this process to learn plans for the game of chess. The plan construction mechanism consists of the following three stages: - Data Compression - Causal Traceback - Feature Abstraction The crucial problems that must be solved in order to construct a plan are: when to construct a plan, how to concisely represent a learning situation, how to represent the relevant underlying causality, and how to generalize the experience so its range of applicablity is appropriately expressed. Once the plans have been constructed, the program can then use than either offensively (in action selection), or defensively (in action rejection).

Paper Details

Date Published: 11 December 1985
PDF: 10 pages
Proc. SPIE 0579, Intelligent Robots and Computer Vision IV, (11 December 1985); doi: 10.1117/12.950842
Show Author Affiliations
James C. Spohrer, Yale University (United States)

Published in SPIE Proceedings Vol. 0579:
Intelligent Robots and Computer Vision IV
David P. Casasent, Editor(s)

© SPIE. Terms of Use
Back to Top
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research
Forgot your username?