[Ml-stat-talks] Wed: Lauren Hannah on Multivariate convex regression

David Mimno mimno at CS.Princeton.EDU
Mon Dec 10 09:18:15 EST 2012

For the last ML talk of the semester we welcome back Princeton PhD Lauren


Lauren Hannah, Columbia University, Dept. of Statistics
CS402, 12/12/12 (!), 12:30

Title: Multivariate convex regression

Abstract: Shape constraints, like convexity, occur in a variety of
inference problems, including geometric programming based circuit design,
options pricing and value function approximation. Current methods for
multivariate regression subject to convexity constraints do not scale to
more than a couple of thousand observations. In this talk, we present a new
computationally efficient, consistent method for multivariate convex
regression using adaptive partitioning with linear partition models. We
apply this method to large scale inference problems and discuss extensions
to objective function and constraint approximation in a convex optimization
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/ml-stat-talks/attachments/20121210/b6a6bca6/attachment.html>

More information about the Ml-stat-talks mailing list