Optimizer Comparison on a Non-Convex Loss Surface | AlgoML Prep