Pretrained transformers achieve the state of the art across tasks in natural language processing, motivating researchers to investigate their inner mechanisms. One common direction is to understand what features are important for prediction. In this paper, we apply information bottlenecks to analyze the attribution of each feature for prediction on a black-box model. We use BERT as the example and evaluate our approach both quantitatively and qualitatively. We show the effectiveness of our method in terms of attribution and the ability to provide insight into how information flows through layers. We demonstrate that our technique outperforms two competitive methods in degradation tests on four datasets. Code is available at https://github.com/bazingagin/IBA.
Inserting Information Bottlenecks for Attribution in Transformers
Information bottlenecks are used to analyze feature importance and information flow in BERT, outperforming competitive methods in degradation tests.
- Year
- 2020
- Venue
- Findings of the Association for Computational Linguistics 2020
- Authors
- 4
- Hosting
- Abstract onlyARXIV-DEFAULT
Cite
Notes
Only stored in your browser.
Attribution
- Abstract & full text
- arxiv.org/abs/2012.13838v2ARXIV-DEFAULT
- TL;DR
- Semantic Scholar