Estimates of covering numbers of convex sets with slowly decaying orthogonal subsets (Q2381815)

The authors present a new lower bound for the covering numbers of symmetric convex precompact sets in the infinite-dimensional Hilbert spaces. The bound is given in terms of decay functions of the nearly orthogonal subsets and shown to be tight in the case of convex hulls of sets with power-type covering numbers. The proofs use special properties of orthogonal subsets of convex sets and the exponential growth of the size of generalized Hadamard matrices. Applications to machine learning, neurocomputing and nonlinear approximation are discussed.