A note on a Lyapunov argument for stochastic gradient methods in the presence of noise