Fetching the paper…

Blockwise Adaptivity: Faster Training and Better Generalization in Deep Learning · Around