Abstract
A recent study by Feldman (2020) proposed a long-tail theory to explain the memorization behavior of deep learning models. However, memorization has not been empirically verified in the context of NLP, a gap addressed by this work. In this paper, we use three different NLP tasks to check if the long-tail theory holds. Our experiments demonstrate that top-ranked memorized training instances are likely atypical, and removing the top-memorized training instances leads to a more serious drop in test accuracy compared with removing training instances randomly. Furthermore, we develop an attribution method to better understand why a training instance is memorized. We empirically show that our memorization attribution method is faithful, and share our interesting finding that the top-memorized parts of a training instance tend to be features negatively correlated with the class label.- Anthology ID:
- 2022.acl-long.434
- Volume:
- Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
- Month:
- May
- Year:
- 2022
- Address:
- Dublin, Ireland
- Venue:
- ACL
- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 6265–6278
- Language:
- URL:
- https://aclanthology.org/2022.acl-long.434
- DOI:
- 10.18653/v1/2022.acl-long.434
- Cite (ACL):
- Xiaosen Zheng and Jing Jiang. 2022. An Empirical Study of Memorization in NLP. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 6265–6278, Dublin, Ireland. Association for Computational Linguistics.
- Cite (Informal):
- An Empirical Study of Memorization in NLP (Zheng & Jiang, ACL 2022)
- PDF:
- https://preview.aclanthology.org/remove-xml-comments/2022.acl-long.434.pdf
- Code
- xszheng2020/memorization
- Data
- CIFAR-10, SST, Yahoo! Answers