Approximation by combinations of ReLU and squared ReLU ridge functions with ℓ¹ and ℓ⁰ controls

We establish L^∞ and L² error bounds for functions of many variables that are approximated by linear combinations of rectified linear unit (ReLU) and squared ReLU ridge functions with ¹ and ⁰ controls on their inner and outer parameters. With the squared ReLU ridge function, we show that the L² approximation error is inversely proportional to the inner layer ⁰ sparsity and it need only be sublinear in the outer layer ℓ⁰ sparsity. Our constructions are obtained using a variant of the Maurey-Jones-Barron probabilistic method, which can be interpreted as either stratified sampling with proportionate allocation or two-stage cluster sampling. We also provide companion error lower bounds that reveal near optimality of our constructions. Despite the sparsity assumptions, we showcase the richness and flexibility of these ridge combinations by defining a large family of functions, in terms of certain spectral conditions, that are particularly well approximated by them.