摘要 |
The invention relates to a method for the computer-aided studying of a neural network (1) comprising a plurality of neurons. According to said method, a) the neurons of the neural network (1) are divided into at least two layers (L1, L2) comprising a first layer (L1) and a second layer (L2) that is linked to the first layer (L1); b) input data (I) in the first layer (L1) is represented by one or several manifestations (D1, D2, N1, N2) of one or several features, each feature manifestation (D1, D2, N1, N2) encompassing one or several neurons of the first layer while a plurality of categories (C1, C2) are stored in the second layer, each category comprising one or several neurons of the second layer (L2); c) at least one category (C1, C2) in the second layer (L2) is allocated to the feature manifestation (D1, D2, N1, N2) of the input data (I) in the first layer (L1) for one or several pieces of input data (I); d) a piece of input data (I) is input into the first layer (L1), whereupon at least one state variable of the neural network (1) is determined and is compared to the at least one category (C1, C2) of said input data (I) allocated in step c); e) the link between the first and the second layer (L1, L2) is modified in accordance with the result of the comparison made in step d). |
申请人 |
SIEMENS AKTIENGESELLSCHAFT;DECO, GUSTAVO;STETTER, MARTIN;SZABO, MIRUNA |
发明人 |
DECO, GUSTAVO;STETTER, MARTIN;SZABO, MIRUNA |