Integrating Machine Learning in Parallel Heuristic Search

Many artificial intelligence applications rely on searching through large, complex spaces. Iterative Deepening-A* (IDA’) is a procedure capable of finding a least cost path to a goal; however, the execution time on a single processor is often too long for most applications. Parallel processing canbe used to reduce the complexity and run time of a search problem by dividing the work load over multiple processors. A number of techniques have been introduced that improve the performance of heuristic search. Each approach has a number of parameters that are often selected bythe user before the problem isexecuted. It is common to find that the selection of parameters for one approach will conflict with parameters for another approach and poor performance will result. This research uses a knowledge base constructed by the C4.5 machine learning system to continuously elect approaches and the associated parameters throughout the execution of a problem. Therefore, as the search moves deeper into the tree structure, the approach will be modified to adapt to the tree structure. A series of experiments has been performed with the fifteen puzzle problem domain. A significant improvement in performance has resulted from using machine learning to select an appropriate approach for a problem.

[1]  Shantanu Dutt,et al.  New Anticipatory Load Balancing Strategies for Parallel A* Algorithms , 1994, Parallel Processing of Discrete Optimization Problems.

[2]  George A. Miller,et al.  WordNet: A Lexical Database for English , 1995, HLT.

[3]  Richard E. Korf,et al.  Depth-First Iterative-Deepening: An Optimal Admissible Tree Search , 1985, Artif. Intell..

[4]  Alberto Maria Segre,et al.  Programs for Machine Learning , 1994 .

[5]  Diane J. Cook,et al.  Parallel search using transformation‐ordering Lterative‐Deepening‐A* , 1993, Int. J. Intell. Syst..

[6]  V. Nageshwara Rao,et al.  Scalable parallel formulations of depth-first search , 1990 .

[7]  Richard E. Korf,et al.  Single-Agent Parallel Window Search , 1991, IEEE Trans. Pattern Anal. Mach. Intell..

[8]  Robert Craig Varnell An architecture for improving the performance of parallel search , 1998 .

[9]  V. A. Saletore A Distributed and Adaptive Dynamic Load Balancing Scheme for Parallel Processing of Medium-Grain Tasks , 1990, Proceedings of the Fifth Distributed Memory Computing Conference, 1990..

[10]  Vipin Kumar,et al.  Scalable Load Balancing Techniques for Parallel Computers , 1994, J. Parallel Distributed Comput..

[11]  Larry S. Davis,et al.  Parallel Iterative A* Search: An Admissible Distributed Heuristic Search Algorithm , 1989, IJCAI.

[12]  J. Ross Quinlan,et al.  C4.5: Programs for Machine Learning , 1992 .

[13]  Sartaj Sahni,et al.  Load balancing on a hypercube , 1991, [1991] Proceedings. The Fifth International Parallel Processing Symposium.