Uniform approximation of Vapnik-Chervonenkis classes (Q1932231)

Let \(({\mathcal{X,S,\mu}})\) be a probability space and \({\mathcal C}\in {\mathcal S}\) a given family of measurable sets. The man result of the paper states that, if \({\mathcal C}\subset {\mathcal S}\) has finite Vapnik-Chervonenkis dimension, then \({\mathcal C}\) is can be finitely approximated for every probability measure \({\mathcal \mu}\) on \(({\mathcal X,S}\)). From this, it follows that separable Vapnik-Chervonenkis classes of sets have finite bracketing numbers. These classes, for every ergodic process, satisfy uniform laws of large numbers. Similar results for Vapnik-Chervonenkis major and Vapnik-Chervonenkis graph families of functions are obtained.