Neural network employing a location addressable memory and method for operating the same

A neural network comprising:

• indirect association code generation means for obtaining first intermediate states (Z.sub.1i) and second intermediate states (Z.sub.2i) linearly separable from given input data (X.sub.i) and given output data (Y.sub.i);
• an input-side single layer perceptron for generating direct associations (X.sub.i, Z.sub.1i) as linearly separable problems with respect to said given input data (X.sub.i) and said first intermediate states (Z.sub.1i);
• an output-side single layer perceptron for generating direct associations (Z.sub.2i,Y.sub.i) as linearly separable problems with respect to the given output data (Y.sub.i) and said second intermediate states (Z.sub.2i);
• learning means for learning said direct associations (X.sub.i), (Z.sub.1i) and (Z.sub.2i, Y.sub.i) by a least mean square learning process;
• and
• a location addressable memory for connecting the first intermediate states (Z.sub.1i) with respect to said input-side single layer perceptron with the second intermediate states (Z.sub.2i) with respect to said output-side single layer perceptron.

A neural network comprising:

• indirect association code generation means for obtaining intermediate states (Z.sub.i) linearly separable from given input data (X.sub.i) and given output data (Y.sub.i);
• a single layer perceptron for providing direct associations (X.sub.i, Z.sub.i) for said given input data (X.sub.i) and said intermediate states (Z.sub.i);
• learning means for learning said direct associations (X.sub.i, Z.sub.i) by a least mean square learning process;
• and
• a location addressable memory for storing the intermediate states (Z.sub.i) from said single layer perceptron as addresses and for storing said given output data (Y.sub.i) as desired output values, correspondingly to said addresses.