Compressing Pattern Databases with Learning

Authors:
Mehdi Samadi;Maryam Siabani;Ariel Felner;Robert Holte
Affiliations:
Department of Computing Science, University of Alberta, Edmonton, Alberta, Canada T6G 2E8, msamadi@cs.ualberta.ca;Electrical and Computer Engineering Department, Isfahan University of Technology, Isfahan, Iran, siabani@ec.iut.ac.ir;Information Systems Engineering Dept., Deutsche Telekom Labs, Ben Gurion University, Beer-Sheva, Israel, felner@bgu.ac.il;Department of Computing Science, University of Alberta, Edmonton, Alberta, Canada T6G 2E8, holte@cs.ualberta.ca
Venue:
Proceedings of the 2008 conference on ECAI 2008: 18th European Conference on Artificial Intelligence
Year:
2008

Citing 8
Cited 2

On the construction of heuristic functions

On the construction of heuristic functions
Neural network design and the complexity of learning

Neural network design and the complexity of learning
Machine learning and data mining

Communications of the ACM
A Framework for Learning in Search-Based Systems

IEEE Transactions on Knowledge and Data Engineering
Maximizing over multiple pattern databases speeds up heuristic search

Artificial Intelligence
Additive pattern database heuristics

Journal of Artificial Intelligence Research
Compressed pattern databases

Journal of Artificial Intelligence Research
Dual lookups in pattern databases

IJCAI'05 Proceedings of the 19th international joint conference on Artificial intelligence

Inconsistent heuristics in theory and practice

Artificial Intelligence
Learning heuristic functions for large state spaces

Artificial Intelligence

Quantified Score

Hi-index	0.00

Visualization

Abstract

A pattern database (PDB) is a heuristic function implemented as a lookup table. It stores the lengths of optimal solutions for instances of subproblems. Most previous PDBs had a distinct entry in the table for each subproblem instance. In this paper we apply learning techniques to compress PDBs by using neural networks and decision trees thereby reducing the amount of memory needed. Experiments on the sliding tile puzzles and the TopSpin puzzle show that our compressed PDBs significantly outperforms both uncompressed PDBs as well as previous compressing methods. Our full compressing system reduced the size of memory needed by a factor of up to 63 at a cost of no more than a factor of 2 in the search effort.