Hostname: page-component-7c8c6479df-r7xzm Total loading time: 0 Render date: 2024-03-19T07:19:03.896Z Has data issue: false hasContentIssue false

A Note on VC-Dimension and Measure of Sets of Reals

Published online by Cambridge University Press:  14 February 2001

SHAI BEN-DAVID
Affiliation:
Computer Science Department, Technion, Haifa 32000, Israel (e-mail: shai@cs.technion.ac.il)
LEONID GURVITS
Affiliation:
NEC Research Institute, Princeton, NJ 08540, USA (e-mail: gurvits@research.nj.nec.com)

Abstract

Vapnik and Chervonenkis proposed in [7] a combinatorial notion of dimension that reflects the ‘combinatorial complexity’ of families of sets. In the three decades that have passed since that paper, this notion – the Vapnik–Chervonenkis dimension (VC-dimension) – has been discovered to be of primal importance in quite a wide variety of topics in both pure mathematics and theoretical computer science.

In this paper we turn our attention to classes with infinite VC-dimension, a realm thrown into one big bag by the usual VC-dimension analysis. We identify three levels of combinatorial complexity of classes with infinite VC-dimension. We show that these levels fall under the set-theoretic definition of σ-ideals (in particular, each of them is closed under countable unions), and that they are all distinct. The first of these levels (i.e., the family of ‘small’ infinite-dimensional classes) coincides with the family of classes which are non-uniformly PAC-learnable.

Maybe the most surprising contribution of this work is the discovery of an intimate relation between the VC-dimension of a class of subsets of the natural numbers and the Lebesgue measure of the set of reals defined when these subsets are viewed as binary representations of real numbers.

As a by-product, our investigation of the VC-dimension-induced ideals over the reals yields a new proper extension of the Lebesgue measure. Another offshoot of this work is a simple result in probability theory, showing that, given any sequence of pairwise independent events, any random event is eventually independent of the members of the sequence.

Type
Research Article
Copyright
2000 Cambridge University Press

Access options

Get access to the full version of this content by using one of the access options below. (Log in options will check for institutional or personal access. Content may require purchase if you do not have access.)