EMNLP 2020教程:《自然语言处理模型的可解释性研究》及相关论文
时间: 2020-11-23 14:48
AMiner平台(https://www.aminer.cn)由清华大学计算机系研发,拥有我国完全自主知识产权。平台包含了超过2.3亿学术论文/专利和1.36亿学者的科技图谱,提供学者评价、专家发现、智能指派、学术地图等科技情报专业化服务。系统2006年上线,吸引了全球220个国家/地区1000多万独立IP访问,数据下载量230万次,年度访问量超过1100万,成为学术搜索和社会网络挖掘研究的重要数据和实验平台。
EMNLP2020第二天,线上分享火热进行中。昨天已经和大家分享了开幕式上的一些关键点:中国接受率仅为13%,可解释性和模型分析成为“新趋势”,接下来几天,我们会陆续和大家分享EMNLP2020上的教程,想要了解作者和论文信息的同学可以移步:https://www.aminer.cn/conf/emnlp2020/homepage 。
尽管神经NLP模型具有很高的表现力并且在诸多任务上取得了成功,但它们往往做出违反直觉的决策,并且在决策过程中不透明。本教程将提供可解释技术的背景知识,重点关注解释NLP模型预测的方法。
讲者将首先在某一些场景下展示了特定于示例的解释。接下来,讲者全面研究了特定于示例的解释,包括显着性图,输入扰动(例如LIME,输入减少),对抗攻击和影响函数。除了这些讲解之外,讲者还将遍历源代码,这些源代码创建并可视化各种NLP任务的解释。
最后,讲者将讨论该领域的未解决问题,例如评估,扩展和改进解释方法。
1.论文标题:Towards Interpretable Reasoning over Paragraph Effects in Situation
作者:Mucheng Ren, Xiubo Geng, Tao Qin, Heyan Huang, Daxin Jiang
论文链接:https://www.aminer.cn/pub/5f7c384791e0117ac2a788b1?conf=emnlp2020
2.论文标题:Learning Variational Word Masks to Improve the Interpretability of Neural Text Classifiers
作者:Hanjie Chen, Yangfeng Ji
论文链接:https://www.aminer.cn/pub/5f7ae3b991e011983cc81d93?conf=emnlp2020
3.论文标题:Interpretation of NLP models through input marginalization
作者:Siwon Kim, Jihun Yi, Eunji Kim, Sungroh Yoon
论文链接:https://www.aminer.cn/pub/5f7fe6d80205f07f6897322f?conf=emnlp2020
4.论文标题:COGS: A Compositional Generalization Challenge Based on Semantic Interpretation
作者:Najoung Kim, Tal Linzen
论文链接:https://www.aminer.cn/pub/5f7fe6d80205f07f6897317d?conf=emnlp2020
5.论文标题:Towards Interpreting BERT for Reading Comprehension Based QA
作者:Sahana Ramnath, Preksha Nema, Deep Sahni, Mitesh M. Khapra
论文链接:https://www.aminer.cn/pub/5f7fe6d80205f07f689733fc?conf=emnlp2020
6.论文标题:How do Decisions Emerge across Layers in Neural Models? Interpretation with Differentiable Masking
作者:De Cao Nicola, Schlichtkrull Michael, Aziz Wilker, Titov Ivan
论文链接:https://www.aminer.cn/pub/5eabf34c91e011664ffd2a22?conf=emnlp2020
7.论文标题:PRover: Proof Generation for Interpretable Reasoning over Rules
作者:Swarnadeep Saha, Sayan Ghosh, Shashank Srivastava, Mohit Bansal
论文链接:https://www.aminer.cn/pub/5f7d9cbe91e011346ad27f27?conf=emnlp2020
8.论文标题:Interpretable Multi dataset Evaluation for Named Entity Recognition
作者:Jinlan Fu, Pengfei Liu, Graham Neubig
论文链接:https://www.aminer.cn/pub/5f7fe6d80205f07f6897322e?conf=emnlp2020
9.论文标题:Multimodal Routing: Improving Local and Global Interpretability of Multimodal Language Analysis
作者:Yao-Hung Hubert Tsai, Martin Ma, Muqiao Yang, Ruslan Salakhutdinov, Louis-Philippe Morency
论文链接:https://www.aminer.cn/pub/5f7fe6d80205f07f68973292?conf=emnlp2020
10.论文标题:Interpreting Open Domain Modifiers: Decomposition of Wikipedia Categories into Disambiguated Property Value Pairs
作者:Marius Pasca
论文链接:https://www.aminer.cn/pub/5f7fe6d80205f07f68973230?conf=emnlp2020
11.论文标题:KERMIT: Complementing Transformer Architectures with Encoders of Explicit Syntactic Interpretations
作者:Fabio Massimo Zanzotto, Andrea Santilli, Leonardo Ranaldi, Dario Onorati, Pierfrancesco Tommasino, Francesca Fallucchi
论文链接:https://www.aminer.cn/pub/5f7fe6d80205f07f68973241?conf=emnlp2020
12.论文标题:Cold Start and Interpretability: Turning Regular Expressions into Trainable Recurrent Neural Networks
作者:Chengyue Jiang, Yinggong Zhao, Shanbo Chu, Libin Shen, Kewei Tu
论文链接:https://www.aminer.cn/pub/5f7fe6d80205f07f6897317f?conf=emnlp2020
教程相关材料:https://github.com/Eric-Wallace/interpretability-tutorial-emnlp2020/
EMNLP2020会议详情页:https://www.aminer.cn/conf/emnlp2020/papers
EMNLP历年数据:https://www.aminer.cn/conference/5eeb1307b5261c744f15bda0
扫码了解更多EMNLP2020会议信息
添加“小脉”微信,留言“EMNLP”,即可加入【EMNLP会议交流群】,与更多论文作者学习交流!