English

Why are Adaptive Methods Good for Attention Models?

Optimization and Control 2020-10-26 v2 Machine Learning

Abstract

While stochastic gradient descent (SGD) is still the \emph{de facto} algorithm in deep learning, adaptive methods like Clipped SGD/Adam have been observed to outperform SGD across important tasks, such as attention models. The settings under which SGD performs poorly in comparison to adaptive methods are not well understood yet. In this paper, we provide empirical and theoretical evidence that a heavy-tailed distribution of the noise in stochastic gradients is one cause of SGD's poor performance. We provide the first tight upper and lower convergence bounds for adaptive gradient methods under heavy-tailed noise. Further, we demonstrate how gradient clipping plays a key role in addressing heavy-tailed gradient noise. Subsequently, we show how clipping can be applied in practice by developing an \emph{adaptive} coordinate-wise clipping algorithm (ACClip) and demonstrate its superior performance on BERT pretraining and finetuning tasks.

Keywords

Cite

@article{arxiv.1912.03194,
  title  = {Why are Adaptive Methods Good for Attention Models?},
  author = {Jingzhao Zhang and Sai Praneeth Karimireddy and Andreas Veit and Seungyeon Kim and Sashank J Reddi and Sanjiv Kumar and Suvrit Sra},
  journal= {arXiv preprint arXiv:1912.03194},
  year   = {2020}
}
R2 v1 2026-06-23T12:38:13.162Z