Nearly Uniform Validation Improves Compression-Based Error Bounds

This paper develops bounds on out-of-sample error rates for support vector machines (SVMs). The bounds are based on the numbers of support vectors in the SVMs rather than on VC dimension. The bounds developed here improve on support vector counting bounds derived using Littlestone and Warmuth's compression-based bounding technique.

• Type: Journal
• Year: 2008
• Where: JMLR
• Authors: Eric Bax
• Keywords: JMLR 2008, Out-of-sample Error Rates, Support Vector, Support Vector Machines