Advances in Large Margin Classifiers
Authors :
Alexander J. Smola,
Peter L. Bartlett,
Bernhard Schölkopf,
Dale Schuurmans
Publisher :
MIT Press, Cambridge
Publication Date : 2000
Book Excerpts:
The concept of Large Margins has recently been identified as a unifying principle for analyzing many different approaches to the problem of learning to classify data from examples, including Boosting,
Mathematical Programming,
Neural Networks and
Support Vector Machines. The fact that it is the margin, or confidence level, of a classification (i.e., a scale parameter) rather than the raw training error that matters has become a key tool in recent years when dealing with classifiers. The present volume shows that this applies both to the theoretical analysis and to the design of algorithms.
Whilst the origin of some of these methods dates back to the work of
Vapnik,
Mangasarian and others in the 1960s, it took until the 1990s until applications on large real-world problems began. This is due to both the computational resources that recently become available, and theoretical advances, for instance regarding the
nonlinear generalization of algorithms. At present, algorithms that explicitly or implicitly exploit the concept of margins are among the most promising approaches to learning from data.
A two-day workshop on this topic was organized at the annual
Neural Information Processing Systems (NIPS) conference, held in Breckenridge, Colorado, in December 1998. The present volume contains a number of papers based on talks presented at the workshop along with a few articles describing results obtained since the workshop has taken place. Although it is far too early to give a final analysis of Large Margin Classifiers, this book attempts to provide a first overview of the subject. Hopefully, it will help making large margin techniques part of the standard toolbox in data analysis and prediction, and that it will serve as a starting point for further research.
View/Download Advances in Large Margin Classifiers -- URL's no longer available
