How Does Fine-tuning Affect the Geometry of Embedding Space: A Case Study on Isotropy

Sara Rajaee, Mohammad Taher Pilehvar


Abstract
It is widely accepted that fine-tuning pre-trained language models usually brings about performance improvements in downstream tasks. However, there are limited studies on the reasons behind this effectiveness, particularly from the viewpoint of structural changes in the embedding space. Trying to fill this gap, in this paper, we analyze the extent to which the isotropy of the embedding space changes after fine-tuning. We demonstrate that, even though isotropy is a desirable geometrical property, fine-tuning does not necessarily result in isotropy enhancements. Moreover, local structures in pre-trained contextual word representations (CWRs), such as those encoding token types or frequency, undergo a massive change during fine-tuning. Our experiments show dramatic growth in the number of elongated directions in the embedding space, which, in contrast to pre-trained CWRs, carry the essential linguistic knowledge in the fine-tuned embedding space, making existing isotropy enhancement methods ineffective.
Anthology ID:
2021.findings-emnlp.261
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2021
Month:
November
Year:
2021
Address:
Punta Cana, Dominican Republic
Editors:
Marie-Francine Moens, Xuanjing Huang, Lucia Specia, Scott Wen-tau Yih
Venue:
Findings
SIG:
SIGDAT
Publisher:
Association for Computational Linguistics
Note:
Pages:
3042–3049
Language:
URL:
https://aclanthology.org/2021.findings-emnlp.261
DOI:
10.18653/v1/2021.findings-emnlp.261
Bibkey:
Cite (ACL):
Sara Rajaee and Mohammad Taher Pilehvar. 2021. How Does Fine-tuning Affect the Geometry of Embedding Space: A Case Study on Isotropy. In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 3042–3049, Punta Cana, Dominican Republic. Association for Computational Linguistics.
Cite (Informal):
How Does Fine-tuning Affect the Geometry of Embedding Space: A Case Study on Isotropy (Rajaee & Pilehvar, Findings 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.findings-emnlp.261.pdf
Video:
 https://aclanthology.org/2021.findings-emnlp.261.mp4