Size and depth of monotone neural networks: interpolation and approximation

Dan Mikulincer, Daniel Reichman

We study monotone neural networks with threshold gates where all the weights (other than the biases) are non-negative. We focus on the expressive power and efficiency of representa…