Title :
Progressive stochastic search for solving constraint satisfaction problems
Author :
Lam, Bryan Chi-ho ; Leung, Ho-fung
Author_Institution :
Dept. of Comput. Sci. & Eng., Chinese Univ. of Hong Kong, China
Abstract :
Stochastic search methods have attracted much attention of the constraint satisfaction problem (CSP) research community. Traditionally, a stochastic solver escapes from local optima or leaves plateaus by random restart or heuristic learning. In this paper, we propose the progressive stochastic search (PSS) and its variants for solving binary CSPs, in which a variable always has to choose a new value when it is designated to be repaired. Intuitively, the search can be thought to be mainly driven by a "force" to "rush through" the local minima and plateaus. Timing results show that this approach significantly outperforms LSDL(GENET) (Choi et al, 2000) in N-Queens, Latin squares, random permutation generation problems and randomly CSPs, while it fails to win LSDL(GENET) in quasigroup completion problems and increasing permutation generation problems. This prompts an interesting new research direction in the design of stochastic search schemes.
Keywords :
computability; constraint theory; problem solving; search problems; stochastic processes; LSDL (GENET); Latin squares; N-Queens; binary CSP; constraint satisfaction problem; heuristic learning; progressive stochastic search; random CSP; random permutation generation problem; stochastic solver; Artificial intelligence; Computer science; Cost function; Search methods; Stochastic processes; Timing; Tin;
Conference_Titel :
Tools with Artificial Intelligence, 2003. Proceedings. 15th IEEE International Conference on
Print_ISBN :
0-7695-2038-3
DOI :
10.1109/TAI.2003.1250229