 |
TITLE: Estimating sparse models from multivariate discrete data via transformed Lasso
SPEAKER: Teemu Roos (Helsinki Institute for Information Technology HIIT, Finland)
DATE: 2:00 - 3:00 PM, Tuesday, February 10, 2009
LOCATION: Tioga, 3U
ABSTRACT:
The type of L1 norm regularization used in Lasso and related methods typically
yields sparse parameter estimates where most of the estimates are equal to
zero. We study a class of estimators obtained by applying a linear
transformation on the parameter vector before evaluating the L1 norm. The
resulting "transformed Lasso" yields estimates that are "smooth" in a way that
depends on the applied transformation. The optimization problem is convex and
can be solved efficiently using existing tools. We present two examples: the
Haar transform which corresponds to variable length Markov chain (context-tree)
models, and the Walsh-Hadamard transform which corresponds to linear
combinations of XOR (parity) functions of binary input features.
BIOGRAPHY:
Teemu Roos is a postdoctoral researcher with the Helsinki Institute for
Information Technology HIIT, Finland. He obtained his MSc and PhD (2007) in
computer science at the University of Helsinki. His research interests include
Bayesian and information-theoretic approaches to data analysis and machine
learning.
|
|
|
printable version
