Automatic Construction of a Hierarchical Generate-and-Test Algorithm

Sunil Mohan, Chris Tong

Abstract: Publisher Summary This chapter describes research on automating the construction of such hierarchical solvers. Hierarchical problem solvers prune out large chunks of the search space by first constructing an abstract solution and then refining this into a concrete solution to the given problem. The chapter presents a procedure called HiT that builds a hierarchical Generate and Test (G&T) problem solver by constructing an abstract search space and problem space, related to the corresponding spaces of the given problem domain through an abstraction function. The resulting system is guaranteed to be complete and satisfies Tenenberg's upward solution property. HiT builds a new abstract CSP domain on top of the provided (base level) problem domain. The hierarchical system constructed by HiT achieves improved performance by reducing the cost of solving some of the tests of the base level problem. HiT creates an abstract space by clustering the base search space objects.