Orthogonal Least Squares Regression: An Efficient Approach for Parsimonious Modelling from Large Data
- Speaker: Prof Sheng Chen (University of Southampton)
- Host: Ke Chen
- 10th February 2010 at 14:15 in Lecture Theatre 1.4, Kilburn Building
The orthogonal least squares (OLS) algorithm, developed in the late 1980s for nonlinear system modelling, remains highly popular for nonlinear data modelling practicians, for the reason that the algorithm is simple and efficient, and is capable of producing parsimonious nonlinear models with good generalisation performance. Since its derivation, many enhanced variants of the OLS forward regression have been developed by incorporating the recent new developments from machine learning. Notably, regularisation techniques, optimal experimental design methods and leave-one-out cross validation have been combined with the OLS algorithm. The resultant class of OLS algorithms offers the state-of-the-art for parsimonious modelling from large data. Other topics discussed in this talk include effective grey-box modelling by incorporating the prior knowledge naturally to the model structure, and further efficiency enhancement for the OLS forward regression modelling by implementing the branch and bound strategy with the OLS algorithm.