LINEAR ASSOCIATIVE MAP COMPUTING ELEMENT AND OPTIMAL LINEAR ASSOCIATIVE MAP LEARNING METHOD USING THE SAME

摘要

PURPOSE:To design a computing element consisting of a simple circuit without requiring a multi-input adder by obtaining the value of a current source as the product of a linear map input value and a weight, and obtaining the value of the linear map as the sum of the potentials of the current source of a boundary part. CONSTITUTION:Current sources J are many as the number of the dimension of the input vector of the linear map are arranged in a grating shape, and one terminals (t) are connected through resistance lines at a constant resistance value (r). Further the current sources at the boundary part are grounded through the resistance lines at the same resistance value (r) and the other terminals of the all current sources J are grounded. In such a circuit, the current value in the direction from one terminal to the other terminal of the current source is obtained as the product of the linear map input vector and the weight vector, and the linear map value can be obtained as the sum of the potentials at one edge of the current source at the boundary part. Thus the simple grating type network circuit can be used without using the multi-input adder, and the design can be facilitated.