摘要 |
PURPOSE:To obtain a neural network with small number of connections between units and easy to make into a device even when an input pattern is a complicated one and also, the number of identification categories is increased. CONSTITUTION:The number of layers is set as the number >=4, and an inter-unit coupling load value between the unit included in an input layer and the unit included in a first intermediate layer neighboring to the input layer is fixed, and an output value from each unit in the first intermediate layer is expressed in a value relating to the mutual correlation of a load pattern in which the coupling load values from an input layer unit to the unit are arranged and an input pattern in which the output value from each unit in the input layer is arranged and also, in an analog value or multi-digital value over a ternary value, and also, the absolute value of the value is set as a definite value, and each coupling load value between each unit from the first intermediate layer to an output layer can be decided by learning with teacher.
|