An Experimental Study on Asymptotic Learning Curves

スポンサーリンク

概要

• 論文の詳細を見る

• Stochastic multi-layer feed-forward networks are trained with back-propagation and the learning curves are compared to previous work by Amari et al. The asymptotic generalization error scales as 1/t as predicted (t is the number of training patterns). For an intermediate range a faster 1/t^2 scaling is observed. This effect can explained by using higher order corrections of the likelihood expansion.

• 一般社団法人電子情報通信学会の論文
• 1995-06-30

著者

• Muller K.-r.

  Dept. Of Math. Engineering And Inf. Physics University Of Tokyo:gmd First:beckman Institute Universi

• Finke M.

  Institut fur Logik, University of Karlsruhe

• Murata N.

  Dept. of Math. Engineering and Inf. Physics, University of Tokyo

• Schulten K.

  Beckman Institute, University of Illinois

• Amari S.

  Dept. of Math. Engineering and Inf. Physics, University of Tokyo

• Finke M.

  Institut Fur Logik University Of Karlsruhe

• Schulten K.

  Beckman Institute University Of Illinois

関連論文

• An Experimental Study on Asymptotic Learning Curves

スポンサーリンク