Bayesian Optimization Using Hamiltonian Dynamics for Sparse Artificial Neural Networks