Novel relearning algorithm of neural network
MA Ruiping · Systems engineering and electronics · 2005
Obtaining training specimen is a bottleneck in the application of NN. The existing method, which is to rebuild and retrain NN when a new training specimen appears, has two weaknesses: (1) the time of training is increased; (2) old memory is affected. A new method is proposed that a local NN is added based on the eriginal NN. This local NN only includes new specimen mode and is isolated with the original NN. Only local NN is trained when the NN is relearning.