CONSTITUTING METHOD, LEARNING METHOD, AND GENERALIZING ABILITY EVALUATING METHOD FOR NEURAL NETWORK

Japanese Patent JP3304476

B2

PURPOSE: To constitute a neural network which is provided with an input layer neuron unit, an intermediate layer neuron unit, and an output layer neuron unit, and has excellent generalizing ability as the result of learning by plural kinds of input signals related to a recognition category.
CONSTITUTION: The input layer, the intermediate layer, and the output layer neutron units (SA1-SA64, AA1-AA5, RA1-RA16) are arranged so that each input/output function f (x) establishes 0<f (x) ≤1, and the learning by the technique of back-propagation is executed under a condition that the initial value of inter-input and intermediate layer synapse transmission efficiency is set so that the output of the intermediate layer neutron unit is made the lowest level when all the inputs to the input layer neuron unit are the lowest level, and simultaneously, the initial value of synapse transmission efficiency between intermediate and output layer is set so that the output of the output layer neutron unit is made the highest level when each output of the intermediate layer neutron unit is the lowest level, and the value of the synapse transmission efficiency between input and intermediate layer and the value of the synapse transmission efficiency between intermediate and output layer are determined.

Akimasa Shirosaka

JP5408093A

July 22, 2002

March 15, 1993

Click for automatic bibliography generation  

Mazda Corporation Inc

G06F15/18; G06G7/60; G06N3/04; G06N3/08; G06N99/00; (IPC1-7): G06N3/08; G06N3/04

JP3201159A

二木徹，「識別問題におけるバックプロパゲーションアルゴリズムの初期値決定法」，電気学会論文誌Ｃ，日本，社団法人電気学会・発行，１９９０年３月２０日，Ｖｏｌ．１１０，Ｎｏ．３，ｐｐ．１１９−１２５，特許庁ＣＳＤＢ文献番号：ＣＳＮＴ１９９９００３０３００２

Sadaaki Kanbara

Previous Patent: RESIN COMPOSITION WHICH INDICATES DRUG DEFICACY MEDICINE

Next Patent: PRODUCTION OF ALPHA, BETA-SUBSTITUTED CYCLOPENTANONE DERIVATIVE