234

Fat-shattering dimension of kk-fold maxima

Journal of machine learning research (JMLR), 2021
Abstract

We provide improved estimates on the fat-shattering dimension of the kk-fold maximum of real-valued function classes. The latter consists of all ways of choosing kk functions, one from each of the kk classes, and computing their pointwise maximum. The bound is stated in terms of the fat-shattering dimensions of the component classes. For linear and affine function classes, we provide a considerably sharper upper bound and a matching lower bound, achieving, in particular, an optimal dependence on kk. Along the way, we point out and correct a number of erroneous claims in the literature.

View on arXiv
Comments on this paper