DMCA
How long will it take (1992)
| Venue: | In Proc. of the 10th National Conf. on AI, 466–471. AAAI |
| Citations: | 11 - 0 self |
BibTeX
@INPROCEEDINGS{Musick92howlong,
author = {Ron Musick and Stuart Russell},
title = {How long will it take},
booktitle = {In Proc. of the 10th National Conf. on AI, 466–471. AAAI},
year = {1992}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
We present a method for approximating the ex-pected number of steps required by a heuristic search algorithm to reach a goal from any ini-tial state in a problem space. The method is based on a mapping from the original state space to an abstract space in which states are charac-terized only by a syntactic “distance ” from the nearest goal. Modeling the search algorithm as a Markov process in the abstract space yields a simple system of equations for the solution time for each state. We derive some insight into the behavior of search algorithms by examining some closed form solutions for these equations; we also show that many problem spaces have a clearly de-lineated “easy zone”, inside which problems are trivial and outside which problems are impossible. The theory is borne out by experiments with both Markov and non-Markov search algorithms. Our results also bear on recent experimental data sug-gesting that heuristic repair algorithms can solve large constraint satisfaction problems easily, given a preprocessor that generates a sufficiently good initial state.
Keyphrases
search algorithm non-markov search algorithm abstract space yield syntactic distance ini-tial state solution time abstract space many problem space de-lineated easy zone ex-pected number original state space heuristic repair algorithm closed form solution simple system heuristic search algorithm problem space recent experimental data sug-gesting good initial state large constraint satisfaction problem markov process
