arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2012.13838cs.CLcs.LG
收藏

在Transformers中插入属性的信息瓶颈

Inserting Information Bottlenecks for Attribution in Transformers

University of Waterloo(滑铁卢大学)

Zhiying Jiang, Raphael Tang, Ji Xin, Jimmy Lin

摘要:

预先训练的Transformers在自然语言处理的任务中达到了最先进的水平,激励研究人员研究其内在机制。一个共同的方向是了解哪些特征对预测很重要。在本文中,我们应用信息瓶颈来分析黑盒模型预测的每个特征的属性。我们以BERT为例,定量和定性地评估我们的方法。我们展示了我们的方法在归因方面的有效性,并能够深入了解信息如何通过层流动。我们证明了我们的技术优于两个竞争力的方法在四个数据集上的退化测试。代码可在https://github.com/bazingagin/IBA上获得。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑