A Method of Learning Automata for Graph Partitioning Problems

A Method of Learning Automata for Graph Partitioning Problems Fei Qian, Member (Hiroshima Kokusai Dakuin University), Shigeya Ikebou, Student Member (Hiroshima Kokusai Dakuin Universityy), Hironori Hirata, Member (Chiba University) Graph partitioning problems (GPP) are the most important NP complete problems faced on the design of VLSI chips. Since one of the most powerful methods to solve such problems is simulated annealing (SA) algorithm, it wastes too much time in annealing process. In this paper, we propose a method of learning automata for GPP and construct two algorithms based on the learning automata with a fixed structure and the learning automata with a variable structure . The computer simulations show that our method has comparable accuracy to SA and faster convergent speed than SA. It also shows that our method has better accuracy than mean field algorithm (MFA) and comparable convergent speed to MFA.