Hamiltonian Monte Carlo on ReLU Neural Networks is Inefficient

We analyze the error rates of the Hamiltonian Monte Carlo algorithm with leapfrog integrator for Bayesian neural network inference.