TY - JOUR
T1 - Fat-Shattering Dimension of k-fold Aggregations
AU - Kontorovich, Aryeh
AU - Attias, Idan
PY - 2024/5
Y1 - 2024/5
N2 - We provide improved estimates on the fat-shattering dimension of the k-fold maximum of real-valued function classes. The latter consists of all ways of choosing k functions, one from each of the k 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 k. Along the way, we point out and correct a number of erroneous claims in the literature.
AB - We provide improved estimates on the fat-shattering dimension of the k-fold maximum of real-valued function classes. The latter consists of all ways of choosing k functions, one from each of the k 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 k. Along the way, we point out and correct a number of erroneous claims in the literature.
U2 - 10.48550/arXiv.2110.04763
DO - 10.48550/arXiv.2110.04763
M3 - Article
SN - 1532-4435
VL - 25
JO - Journal of Machine Learning Research
JF - Journal of Machine Learning Research
ER -