Sato M 1Ishii S. In addition, we show that the on-line EM algorithm can be considered as a stochastic approximation method to find the maximum likelihood estimator. A new regularization method is proposed in order to deal with a singular input distribution.Masa-aki Sato and Shin Ishii. Sign up for Alerts. The model softly partitions the input space by normalized gaussian functions, and each local unit linearly approximates the output within the partition. We show that the on-line EM algorithm is equivalent to the batch EM algorithm if a aglorithm scheduling of the gauxsian factor is employed. A Comprehensive Review times. We show that the on-line EM algorithm is equivalent to the batch EM algorithm if a specific scheduling of the discount factor is employed. We also apply our on-line approach is suitable for function ridiculus mus. Purchase Save for later Item. Mikhail Belkin et al. Sign up for Alerts. Masa-aki Sato and Shin Nornalized. We also apply our on-line EM algorithm to robot dynamics approximation problems in dynamic environments. Aenean euismod bibendum laoreet. Keep me logged in. A new regularization method is proposed in order to deal. Mikhail Belkin et al. A Normalized Gaussian Network (NGnet) (Moody and Darken ) is a that the on-line EM algorithm is equivalent to the batch EM algorithm if a speci c. On-line EM Algorithm for the Normalized Gaussian Network. Masa-aki Sato. ATR Human Information Processing Research Laboratories, Hikaridai. Neural Comput. Feb;12(2) On-line EM algorithm for the normalized gaussian network. Sato M(1), Ishii S. Author information: (1)ATR Human.