Inserting Information Bottlenecks for Attribution in Transformers
This work addresses the problem of understanding feature importance in large transformer models, which is crucial for interpretability in NLP.
This paper applies information bottlenecks to analyze feature attribution in black-box transformer models like BERT. The method effectively attributes features and provides insights into information flow, outperforming two competitive methods in degradation tests across four datasets.
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.