Speaker: Professor Kanta Naito, Department of Mathematics, Shimane University, Japan

Time: 4:00p.m. Wednesday 15th September 2004

Venue: Red Centre Building Room RC-3084,near Barker Street Gate 14

An objective of statistical science is to propose a certain efficient
model describing the underlying structure. The phrase, model", usually
means a mathematical expression of the structure using a finite dimensional
parameter vector. Then we might consider that the role of nonparametric
approach is to adjust the fitting of utilized parametric model. Such
approaches have been realized in smoothing area, called semiparametric
smoothing.

In nonparametric smoothing, data sharpening method has been discussed by
several authors, and it is well known that data sharpening yields reduced bias.

Further, the boosting has become extremely vibrant area in the community of
machine learning, in which the claim that boosting is resistant to overfitting
has been an important problem to be investigated. Among boosting methodologies,
least square boosting is attractive by its tractability.

In this talk, a unified view of above three methods is given in conjunction
with speaker's recent works. All of methods can be seen as adjustment-based
methods, where a smoother of residuals is effectively applied to adjust the
initial fitting.