PLS Mixture Model for Online Dimension Reduction

Authors:
Jiro Hayami;Koichiro Yamauchi
Affiliations:
Graduate School of Information Science and Technology, Hokkaido University, Sapporo Hokkaido, Japan 060-0814;Graduate School of Information Science and Technology, Hokkaido University, Sapporo Hokkaido, Japan 060-0814
Venue:
Neural Information Processing
Year:
2007

Citing 3
Cited 0

Incremental Online Learning in High Dimensions

Neural Computation
On-line EM Algorithm for the Normalized Gaussian Network

Neural Computation
Constructive Incremental Learning from Only Local Information

Neural Computation

Quantified Score

Hi-index	0.00

Visualization

Abstract

This article presents an online learning method for modeling high dimensional input data. This method approximates a nonlinear function by summing up several local linear functions. Each linear function is represented as the weighted sum of a small number of dominant variables, which are extracted by the partial least squares (PLS) regression method. Moreover, a radial function, which represents the respective input area of each linear function, is also redefined using the dominant variables. This article also presents an online deterministic annealing expectation maximization (DAEM) algorithm which includes a temperature control mechanism for acquireing the most suitable system parameters. Experimental results show the effective learning behavior of the new method.