H-Transformer-1D: Fast One-Dimensional Hierarchical Attention for Sequences

Zhenhai Zhu, Radu Soricut


Abstract
We describe an efficient hierarchical method to compute attention in the Transformer architecture. The proposed attention mechanism exploits a matrix structure similar to the Hierarchical Matrix (H-Matrix) developed by the numerical analysis community, and has linear run time and memory complexity. We perform extensive experiments to show that the inductive bias embodied by our hierarchical attention is effective in capturing the hierarchical structure in the sequences typical for natural language and vision tasks. Our method is superior to alternative sub-quadratic proposals by over +6 points on average on the Long Range Arena benchmark. It also sets a new SOTA test perplexity on One-Billion Word dataset with 5x fewer model parameters than that of the previous-best Transformer-based models.
Anthology ID:
2021.acl-long.294
Volume:
Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)
Month:
August
Year:
2021
Address:
Online
Editors:
Chengqing Zong, Fei Xia, Wenjie Li, Roberto Navigli
Venues:
ACL | IJCNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
3801–3815
Language:
URL:
https://aclanthology.org/2021.acl-long.294
DOI:
10.18653/v1/2021.acl-long.294
Bibkey:
Cite (ACL):
Zhenhai Zhu and Radu Soricut. 2021. H-Transformer-1D: Fast One-Dimensional Hierarchical Attention for Sequences. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 3801–3815, Online. Association for Computational Linguistics.
Cite (Informal):
H-Transformer-1D: Fast One-Dimensional Hierarchical Attention for Sequences (Zhu & Soricut, ACL-IJCNLP 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.acl-long.294.pdf
Video:
 https://aclanthology.org/2021.acl-long.294.mp4
Code
 additional community code
Data
Billion Word BenchmarkLRA