Learning Simon's quantum algorithm

Abstract

We consider whether trainable quantum unitaries can be used to discover quantum speed-ups for classical problems. Using methods recently developed for training quantum neural nets, we consider Simon's problem, for which there is a known quantum algorithm which performs exponentially faster in the number of bits, relative to the best known classical algorithm. We give the problem to a randomly chosen but trainable unitary circuit, and find that the training recovers Simon's algorithm as hoped.