Minimization Learning of Neural Networks by Adding Hidden Units
Koichi Niijima, 耕一 新島, Masaaki Yamada, Marghny Hassan Mohamed, 大我 赤沼, Teruya Minamoto, 晃弥 皆本, Akito Ohkubo, 彰人 大久保, Akito Okubo · QIR (Kyushu University Institutional Repository) (Kyushu University) · 1997
This paper proposes a minimization learning method for neural networks with one hidden layer. We treat two types of networks without and with thresholds in their output layers. Both of them are learnt by minimizing error functions whenever one unit is added to the hidden layer. Our learning method is applied to design a neural network with thresholds for image recognition.