Riemannian stochastic optimization methods avoid strict saddle points

Informally, our main result may be stated as follows: Under any stochastic Riemannian Robbins-Monro method, the probability of converging to a strict saddle point (or a submanifold thereof) is zero.