| 摘要 |
PROBLEM TO BE SOLVED: To provide a physical neural network for modeling structure of a complicated physical system, performing training by using incomplete input data, and generating an output desired from the system to an input over a wide-range. SOLUTION: This physical neural network is provided with a plurality of local coordinate nodes 29, and each node 29 has a transfer function for receiving at least one input and converting the input into an output. The output forms one of one or more inputs to another node 29 among the nodes 29. The physical neural network is provided with a set of training nodes having an input to one of a plurality of nodes 29, a set of input nodes having inputs to a plurality of nodes, and training algorithm for regulating a plurality of nodes. At least one transfer function is different from the others, and one of a plurality of nodes is PBNN. COPYRIGHT: (C)2004,JPO
|
