Stochasticity of Deterministic Gradient Descent: Large Learning Rate for Multiscale Objective Function | Read Paper on Bytez