Adaptive Gradient Methods Can Be Provably Faster than SGD after Finite Epochs

ODL
Re-assign community