Neural network parallel computing for optimization problems
Yoshiyasu Takefuji, Kuo-Chun Lee · 1991
This dissertation presents new parallel computing schemes for solving optimization problems based on the artificial neural network. Optimization problems are generally classified as two types of problems: constraint-satisfied problems and minimization problems. A motion equation approach is provided to solve the constraint-satisfied problems more directly. The demonstrated applications for the constraint-satisfied problems are: (1) four-coloring problems; (2) sorting problems; (3) knight's tour problems and others. The proposed network, the generalized maximum neural network, has the following advantages over the existing models: (1) no tuning parameters are required; (2) no threshold value is needed; (3) the equilibrium state is exactly defined; and (4) a feasible solution is always guaranteed. Several demonstrated applications for the minimization problems are as follows: (1) max cut problems; (2) module orientation problems; (3) maximum clique problems. The analog circuit of the proposed network is also presented.