Not necessarily; these learning things are super non convex and it's not clear to me "where" they max out