Completely quantum neural networks

Artificial neural networks are at the heart of modern deep learning algorithms. We describe how to embed and train a general network in quantum annealer without introducing any classical element training. To implement on state-of-the-art annealer, we develop three crucial ingredients: binary encoding free parameters network, polynomial approximation activation function, reduction higher-order polynomials into quadratic ones. Together, these ideas allow loss function as an Ising model Hamiltonian. The then trains by finding ground state. this for elementary illustrate advantages training: its consistency global minimum fact that training converges single annealing step, which leads short times while maintaining high classification performance. Our approach opens novel avenue machine models.