| 摘要 |
<p>PURPOSE:To obtain an optimum solution of the assignment trouble by using a network where neural elements whose number is equal to the number of modules are mutually coupled and setting a coupling coefficient between neural elements to start the network and changing the coupling coefficient value to start it at each time of detecting the convergence of output. CONSTITUTION:Neural elements 1 whose number corresponds to the number of n- number of modules are mutually coupled with coupling coefficient parts 2 among them, and the output value is multiplied by a weight Wij, and the weighted sum of the output value of each neural element and the weight Wij is calculated. The weighted sum and an external input Ix from a terminal 3 are subjected to threshold operation to determine the output value. This operation is repeated until outputs of elements 1a to 1n are stabilized to '1' or '-1' and the module corresponding to each element is divided into two. When a convergence discriminating part 4 detects that all neural elements are converged to the balanced state (divided into two), two-divided results of the output are stored in a memory, and the neural network is started again and a weight calculating part 6 is used to set a new coupling coefficient to coupling coefficient parts 2 in accordance with convergence results of the memory 5. By this constitution, the simple neural network is used to perform effective assignment to cells of plural modules.</p> |
