PURPOSE: To eliminate the need of training in all patterns by connecting an ART module and a target output pattern, executing mapping between recognition classification and the target output pattern, triggering a warning test and deciding closeness between the recognition classification and the target output pattern in the test.
CONSTITUTION: A correction ARTMAP circuit network 200 contains the ART module 225 receiving an input vector 210 and uses the ART two neural circuit network. The ART module 225 is connected to a mapping field 230 receiving the target output pattern 220 as input. The field 230 executes mapping between recognition classification from the ART module 225 and the target output pattern 220. The mapping field 230 triggers the warning test 240 deciding closeness between the recognition classification and the target pattern 220. During the training of the correction ARTMAP circuit network 200, the input pattern 210 and the desired output pattern 220 are outputted to the circuit network 200.
SHIYUI SHIYUN RIN
JIERARUDO EMU NATSUPU
SUU PIN WAN
Next Patent: DATA SORTING DEVICE