Trading Convexity for Scalability
Download
Ronan Collobert, Jason Weston and Léon Bottou. Trading Convexity for Scalability. Proceedings of the Twenty-third International Conference on Machine Learning (ICML 2006), pages 201-208, 2006.
Abstract
Convex learning algorithms, such as Support Vector Machines (SVMs), are often seen as highly desirable because they offer strong practical properties and are amenable to theoretical analysis. However, in this work we show how non-convexity can provide scalability advantages over convexity. We show how concave-convex programming can be applied to produce (i) faster SVMs where training errors are no longer support vectors, and (ii) much faster Transductive SVMs.
BibTeX
@inproceedings{collobert:2006,
author = {R. Collobert and F. Sinz and J. Weston and L. Bottou},
title = {Trading convexity for scalability},
booktitle = {Proceedings of the Twenty-third International Conference on Machine Learning (ICML 2006)},
year = {2006},
pages = {201--208},
location = {Pittsburgh, Pennsylvania},
publisher = {ACM Press}
}
Notes
This paper received the best paper award at ICML 2006 conference.
Links
- The
UniverSVMsoftware implements algorithms in this paper, based on the SVQP2 optimizer. - DC algorithms
Last modified on Tue Apr 15 16:57:43 2008