New (and my first) blog article on the connections between policy gradient and zeroth-order methods posted in "Blogs and Discussion"; check it out.