A structural learning by adding independent noises to hidden units
Takio Kurita, Hideki Asoh, Shinji Umeyama, Shotaro Akaho, Akitaka Hosomi · 1994
The paper demonstrates that a skeletal structure of a network emerges when independent noises are added to the inputs of the hidden units of multilayer perceptron during the learning by error backpropagation. By analyzing the average behavior of the error backpropagation algorithm to such noises, it is shown that the weights from the hidden units to the output units tend to get smaller and the outputs of the hidden units tend to be 0 or 1. Such tendency have been demonstrated by experiments of learning of pattern classification problem.>