Approximation of functions with one-bit neural networks

C. Sinan Güntürk, Weilin Li

The celebrated universal approximation theorems for neural networks roughly state that any reasonable function can be arbitrarily well-approximated by a network whose parameters ar…