Ionel Popescu
University of Bucharest & Simion Stoilow Institute of Mathematics of the Romanian Academy, Bucharest, Romania

One of the problems in neural networks is that they have good intepolation powers. We show that one can interpolate with shallow neural networks as long as the activation function is not a polynomial of $d − 2$ where s is the size of the data. Moreover this purely existential result is complemented by a constructed one where the size of the hidden layer is of size $d · log(d)$. Joint work with Vlad Raul Constantinescu.