Learning Provably Improves the Convergence of Gradient Descent | Read Paper on Bytez