Heavy-Tailed Class Imbalance and Why Adam Outperforms Gradient Descent on Language Models

arXiv:2402.19449v1 Announce Type: new
Abstract: Adam has been shown to outperform gradient descent in optimizing large language transformers empirically, and by a larger margin than on other tasks, but it is unclear why this happens. We show that the heavy-tailed class imbalance found in language modeling tasks leads to difficulties in the optimization dynamics. When training with gradient descent, the loss associated with infrequent words decreases slower than the loss associated with frequent ones. As most samples come from relatively …

abstract adam arxiv class cs.cl cs.lg found gradient language language models large language leads math.oc modeling show stat.ml tasks transformers type