Knowledge Distillation with Noisy Labels for Natural Language Understanding

Shivendra Bhardwaj, Abbas Ghaddar, Ahmad Rashid, Khalil Bibi, Chengyang Li, Ali Ghodsi, Phillippe Langlais, Mehdi Rezagholizadeh


Abstract
Knowledge Distillation (KD) is extensively used to compress and deploy large pre-trained language models on edge devices for real-world applications. However, one neglected area of research is the impact of noisy (corrupted) labels on KD. We present, to the best of our knowledge, the first study on KD with noisy labels in Natural Language Understanding (NLU). We document the scope of the problem and present two methods to mitigate the impact of label noise. Experiments on the GLUE benchmark show that our methods are effective even under high noise levels. Nevertheless, our results indicate that more research is necessary to cope with label noise under the KD.
Anthology ID:
2021.wnut-1.33
Volume:
Proceedings of the Seventh Workshop on Noisy User-generated Text (W-NUT 2021)
Month:
November
Year:
2021
Address:
Online
Venues:
EMNLP | WNUT
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
297–303
Language:
URL:
https://aclanthology.org/2021.wnut-1.33
DOI:
10.18653/v1/2021.wnut-1.33
Bibkey:
Cite (ACL):
Shivendra Bhardwaj, Abbas Ghaddar, Ahmad Rashid, Khalil Bibi, Chengyang Li, Ali Ghodsi, Phillippe Langlais, and Mehdi Rezagholizadeh. 2021. Knowledge Distillation with Noisy Labels for Natural Language Understanding. In Proceedings of the Seventh Workshop on Noisy User-generated Text (W-NUT 2021), pages 297–303, Online. Association for Computational Linguistics.
Cite (Informal):
Knowledge Distillation with Noisy Labels for Natural Language Understanding (Bhardwaj et al., WNUT 2021)
Copy Citation:
PDF:
https://preview.aclanthology.org/update-css-js/2021.wnut-1.33.pdf
Data
GLUEQNLISuperGLUE