1 paper
Foivos Alimisis, Rustem Islamov, Aurelien Lucchi
Learning rate warm-up -- increasing the learning rate at the beginning of training -- has become a ubiquitous heuristic in modern deep learning, yet its theoretical foundations rem…