Review for NeurIPS paper: Stochasticity of Deterministic Gradient Descent: Large Learning Rate for Multiscale Objective Function