Parameter-Efficient Transfer Learning with Diff Pruning

Demi Guo, Alexander Rush, Yoon Kim


Abstract
The large size of pretrained networks makes them difficult to deploy for multiple tasks in storage-constrained settings. Diff pruning enables parameter-efficient transfer learning that scales well with new tasks. The approach learns a task-specific “diff” vector that extends the original pretrained parameters. This diff vector is adaptively pruned during training with a differentiable approximation to the L0-norm penalty to encourage sparsity. As the number of tasks increases, diff pruning remains parameter-efficient, as it requires storing only a small diff vector for each task. Since it does not require access to all tasks during training, it is attractive in on-device deployment settings where tasks arrive in stream or even from different providers. Diff pruning can match the performance of finetuned baselines on the GLUE benchmark while only modifying 0.5% of the pretrained model’s parameters per task and scales favorably in comparison to popular pruning approaches.
Anthology ID:
2021.acl-long.378
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:
4884–4896
Language:
URL:
https://aclanthology.org/2021.acl-long.378
DOI:
10.18653/v1/2021.acl-long.378
Bibkey:
Cite (ACL):
Demi Guo, Alexander Rush, and Yoon Kim. 2021. Parameter-Efficient Transfer Learning with Diff Pruning. 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 4884–4896, Online. Association for Computational Linguistics.
Cite (Informal):
Parameter-Efficient Transfer Learning with Diff Pruning (Guo et al., ACL-IJCNLP 2021)
Copy Citation:
PDF:
https://preview.aclanthology.org/nschneid-patch-2/2021.acl-long.378.pdf
Video:
 https://preview.aclanthology.org/nschneid-patch-2/2021.acl-long.378.mp4
Code
 dguo98/DiffPruning +  additional community code
Data
GLUEQNLISQuAD