Fetching the paper…

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