Title :
Modifications of discrete Hopfield neural optimization in maximum clique problem
Author :
Hwang, Doosung ; Fotouhi, Farshad
Author_Institution :
Dept. of Comput. Sci., Wayne State Univ., Detroit, MI, USA
fDate :
6/24/1905 12:00:00 AM
Abstract :
The Hopfield neural optimization has been studied in maximum clique problem. Its drawback with this approach has the tendency to produce locally optimal solutions due to the descent convergence of the energy function. In order to solve maximum clique problems, the discrete Hopfield neural optimization is studied by combining heuristics such as annealing method and scheduled learning rate which can permit the ascent modification. Each neuron is updated in accordance with a hill-climbing modification. The modifications provide a mechanism for escaping local feasible solutions by varying the direction of motion equation of the neurons. The effectiveness of both modifications is shown through various tests on random graphs and DIMACS benchmark graphs in terms of clique size and computation time
Keywords :
Hopfield neural nets; convergence; gradient methods; graph theory; heuristic programming; mathematics computing; optimisation; DIMACS benchmark graphs; annealing; ascent modification; clique size; computation time; descent convergence; discrete Hopfield neural optimization; energy function; heuristics; hill-climbing modification; local feasible solutions; locally optimal solutions; maximum clique problem; maximum clique problems; motion equation; random graphs; scheduled learning rate; Computer science; Cost function; Equations; Heuristic algorithms; Hopfield neural networks; Neural networks; Neurons; Optimization methods; Simulated annealing; Testing;
Conference_Titel :
Neural Networks, 2002. IJCNN '02. Proceedings of the 2002 International Joint Conference on
Conference_Location :
Honolulu, HI
Print_ISBN :
0-7803-7278-6
DOI :
10.1109/IJCNN.2002.1005460
