| 摘要 |
PURPOSE:To easily realize parallel circuit simulation and to obtain a high speed operation corresponding to parallelism. CONSTITUTION:An (n+1)xn-input and n-output neural network is used which consists of n neuron elements each of which has n+1 inputs and outputs a value (y) expressed by y=wixi (w1 is the weight and Xi is the input signal) and where a j-th synapse load wij of the i-th neuron element is changed by the learning equation expressed by Wij<i+1>=wij<i>-yj (wi, n+1 is always one), and a value cij/cii (cij is an element of a matrix C) is inputted to the j-th input xij (1<=j<=n) of the i-th neuron element, and -yi/cij is substituted into xjn+1 to per-form learning so that an objective output value is (0, 0,... 0); and after learning, one is substituted into the i-th input terminal of the i-th neuron element, and 0 is substituted to the other input terminals, and simulation from the simulation start time to the simulation end time is executed. |
