Maximal width learning of binary functions
Martin Anthony and Joel Ratsaby
CDAM research report series Number LSE-CDAM-2006-11, 2006.

## Abstract

This paper concerns learning binary-valued functions defined on $\bbr$, and investigates how a particular type of `regularity' of hypotheses can be used to obtain better generalization error bounds. We derive error bounds that depend on the {\em sample width} (a notion similar to that of sample margin for real-valued functions). This motivates learning algorithms that seek to maximize sample width.

EPrint Type: Article Project Keyword UNSPECIFIED Computational, Information-Theoretic Learning with StatisticsTheory & Algorithms 2743 Martin Anthony 22 November 2006