New concentration inequalities for suprema of empirical processes (Q470061)

The authors prove concentration inequalities for suprema of random vectors, for which their main application is empirical processes. Such concentration inequalities are well known under assumptions of boundedness, or assumptions on the tails of the underlying random variables. The present work proves such inequalities under the weaker assumption of the existence of some moment of an envelope of the underlying process. More precisely, given a set of random variables \(\{Z_i(j):1\leq j\leq N,1\leq i\leq n\}\), concentration inequalities are proved for \[ Z=\max_{1\leq j\leq N}\left|\frac{1}{n}\sum_{i=1}^nZ_i(j)\right|\,, \] under the assumption of the existence of random variables \(\mathcal{E}_1,\ldots,\mathcal{E}_n\) such that \(|Z_i(j)|\leq\mathcal{E}_i\) for all \(1\leq j\leq N\) and \(1\leq i\leq n\), and such that there exist \(p\in[1,\infty)\) and \(M>0\) (independent of \(N\)) with \(\mathbb{E}\mathcal{E}_i^p\leq M^p\) for all \(1\leq i\leq n\). The authors include interesting and useful discussions of other concentration inequalities available for empirical processes, and of the effectiveness of their bounds.