Research breakthrough possible @S-Logix pro@slogix.in

Office Address

  • 2nd Floor, #7a, High School Road, Secretariat Colony Ambattur, Chennai-600053 (Landmark: SRM School) Tamil Nadu, India
  • pro@slogix.in
  • +91- 81240 01111

Social List

Revealing the Dark Secrets of BERT - 2019

Revealing The Dark Secrets Of Bert

Research Area:  Machine Learning

Abstract:

BERT-based architectures currently give state-of-the-art performance on many NLP tasks, but little is known about the exact mechanisms that contribute to its success. In the current work, we focus on the interpretation of self-attention, which is one of the fundamental underlying components of BERT. Using a subset of GLUE tasks and a set of handcrafted features-of-interest, we propose the methodology and carry out a qualitative and quantitative analysis of the information encoded by the individual BERT’s heads. Our findings suggest that there is a limited set of attention patterns that are repeated across different heads, indicating the overall model overparametrization. While different heads consistently use the same attention patterns, they have varying impact on performance across different tasks. We show that manually disabling attention in certain heads leads to a performance improvement over the regular fine-tuned BERT models.

Keywords:  

Author(s) Name:  Olga Kovaleva, Alexey Romanov, Anna Rogers, Anna Rumshisky

Journal name:  

Conferrence name:  Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing

Publisher name:  Association for Computational Linguistics

DOI:  10.18653/v1/D19-1445

Volume Information: