Statistics Seminar: STA 290
Thursday, October 27th at 4.10pm, MSB 1147 (Colloquium Room)
Speaker: Nourredine El Karoui (UC Berkeley)
Title: Some remarks on penalized and robust regression in high-dimension
There is currently some interest in statistics in understanding the behavior of
multivariate statistical methods in the setting where p, the number of
predictors, and n the number of observations are both large.
In this talk, I will discuss several of these problems (including e.g ridge
regression or regularized discriminant analysis) in the asymptotic setting
where p/n has a finite non-zero limit, as p and n tend to infinity.
I will stress the importance of the concentration of measure phenomenon in
understanding these problems, and its implications for the underlying geometry
of the dataset.
The talk will be based on joint work with Holger Koesters (Bielefeld)
and ongoing work with Derek Bean, Peter Bickel, Chingwhay Lim and Bin Yu