Beyond Preserved Accuracy: Evaluating Loyalty and Robustness of BERT Compression

Canwen Xu, Wangchunshu Zhou, Tao Ge, Ke Xu, Julian McAuley, Furu Wei


Abstract
Recent studies on compression of pretrained language models (e.g., BERT) usually use preserved accuracy as the metric for evaluation. In this paper, we propose two new metrics, label loyalty and probability loyalty that measure how closely a compressed model (i.e., student) mimics the original model (i.e., teacher). We also explore the effect of compression with regard to robustness under adversarial attacks. We benchmark quantization, pruning, knowledge distillation and progressive module replacing with loyalty and robustness. By combining multiple compression techniques, we provide a practical strategy to achieve better accuracy, loyalty and robustness.
Anthology ID:
2021.emnlp-main.832
Volume:
Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing
Month:
November
Year:
2021
Address:
Online and Punta Cana, Dominican Republic
Venue:
EMNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
10653–10659
Language:
URL:
https://aclanthology.org/2021.emnlp-main.832
DOI:
10.18653/v1/2021.emnlp-main.832
Bibkey:
Cite (ACL):
Canwen Xu, Wangchunshu Zhou, Tao Ge, Ke Xu, Julian McAuley, and Furu Wei. 2021. Beyond Preserved Accuracy: Evaluating Loyalty and Robustness of BERT Compression. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 10653–10659, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics.
Cite (Informal):
Beyond Preserved Accuracy: Evaluating Loyalty and Robustness of BERT Compression (Xu et al., EMNLP 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.emnlp-main.832.pdf
Video:
 https://aclanthology.org/2021.emnlp-main.832.mp4
Code
 jetrunner/beyond-preserved-accuracy
Data
GLUEMultiNLI