Showing papers on "Admissible heuristic published in 1993"

An Admissible Heuristic Search Algorithm

Abstract: This paper introduces an admissible heuristic search algorithm — Search and Learning Algorithm (SLA*). SLA* is developed from the work presented by Korf in the Learning-Real-Time-Algorithm (LRTA*). We retain the major elements of Korfs work in LRTA*, and improve its performance by incorporating a review component to fully reflect the effect the learning of new heuristic from front states has upon the previous states. The combined strategy of search, learning, and review has enabled this algorithm to accumulate knowledge continuously through guided expansion, and to identify better search directions in any stage of nodes expansion. With the assumption of non-overestimating initial estimates for all nodes to the goal, this algorithm is able to find an optimal solution in a single problem solving trial with good efficiency. We provide a proof for the optimality of the solution.

Generating effective admissible heuristics by abstraction and reconstitution

Abstract: Admissible heuristics are worth discovering because they have desirable properties in various search algorithms. Unfortunately, effective ones--ones that are accurate and efficiently computable--are difficult for humans to discover. One source of admissible heuristics is from abstractions of a problem: the length of a shortest path solution to an abstracted problem is an admissible heuristic for the original problem because the abstraction has certain details removed. However, often too many details have to be abstracted to yield an efficiently computable heuristic, resulting in inaccurate heuristics. This paper describes a method to reconstitute the abstracted details back into the solution to the abstracted problem, thereby boosting accuracy while maintaining admissibility. Our empirical results of applying this paradigm to project scheduling suggest that reconstitution can make a good admissible heuristic even better.

Autonomous parallel heuristic combinatorial search

Abstract: Heuristic search is the process of searching a state space under the guidance of an evaluation function. Most research on parallelizing heuristic search algorithms has emphasized system problems such as load balancing and reduction in memory use. A theoretical analysis of a new autonomous parallel heuristic search algorithm is introduced. Rather than simply dividing the search space among the processors, the processors share information that monitors the progress of the search and use consensus to limit the amount of time spent in expanding nodes that are not on the optimal path. Each processor uses a different admissible heuristic function, and it is shown that the expected number of nodes generated by each processor in the course of the search is reduced by a factor that reflects the consensus among the processors. The asynchronous behavior of the algorithm eliminates synchronization delays. >