Donald Bren School of Information and Computer Sciences
EASe : integrating search with learned episodes
- Author(s): Ruby, David
- Kibler, Dennis
- et al.
Weak methods are insufficient to solve complex problems. Constrained weak methods, like hill-climbing, search too little of the problem space. Unconstrained weak methods, like breadth-first search, are intractable. Fortunately, through the integration of multiple weak methods more powerful problem solvers can be created. We demonstrate that augmenting a weak constrained search method with episodes provides a tractable method for solving a large class of problems. We demonstrate that these episodes can be generated using an unconstrained weak method while solving simple problems from a domain. We provide an analytical model of our approach and empirical results from the logic synthesis domain of VLSI design as well as the classic tile-sliding domain.