A study of repetitive training with fuzzy clustering
H. Asaka, H. Takahashi, M. Sone, N. Ijjima · 1994
The backpropagation algorithm needs some couples of training data and supervised signals. Generally, the more training data there are, the more certain recognition result is obtained. However, a large number of training data does not always get on the training stage. Thus, this paper propose a new repetitive training algorithm based on fuzzy logic. This algorithm modifies the network on the recognition stage so that the network can output more certain recognition result. As the result, it becomes clear that this algorithm is effective for getting more certain recognition result.>