English

Star-Transformer

Computation and Language 2022-04-26 v3

Abstract

Although Transformer has achieved great successes on many NLP tasks, its heavy structure with fully-connected attention connections leads to dependencies on large training data. In this paper, we present Star-Transformer, a lightweight alternative by careful sparsification. To reduce model complexity, we replace the fully-connected structure with a star-shaped topology, in which every two non-adjacent nodes are connected through a shared relay node. Thus, complexity is reduced from quadratic to linear, while preserving capacity to capture both local composition and long-range dependency. The experiments on four tasks (22 datasets) show that Star-Transformer achieved significant improvements against the standard Transformer for the modestly sized datasets.

Cite

@article{arxiv.1902.09113,
  title  = {Star-Transformer},
  author = {Qipeng Guo and Xipeng Qiu and Pengfei Liu and Yunfan Shao and Xiangyang Xue and Zheng Zhang},
  journal= {arXiv preprint arXiv:1902.09113},
  year   = {2022}
}

Comments

Accepted by NAACL 2019

R2 v1 2026-06-23T07:49:36.180Z