多重化砂時計型ネットを用いた広いクラスの曲面によるデータフィッティング Data Fitting by a Broad Class of Surfaces Using Multiplied Bottleneck Networks
Bottleneck networks were employed in order to estimate a low dimensional surface on which the high dimensional data lie. However, the fitting of closed surfaces like a sphere is hard for them. This is essentially due to the fact that general manifolds cannot be expressed by a single coordinate system. To overcome this difficulty, we multiply the bottleneck networks and construct a mixture of experts network, which can treat a broad class of surfaces. A procedure to restrain the premature convergence of the mixture of experts network is also provided.
- 日本神経回路学会誌 = The Brain & neural networks
日本神経回路学会誌 = The Brain & neural networks 5(1), 3-9, 1998-03-05
Japanese Neural Network Society