| Time |
Details |
| 14:00-14:05 |
Welcom Speech
Jianhua Gao |
| 14:05-14:45 |
Controllable Text Generation
Minlie Huang, Associate professor in Tsinghua University show abstract/bio hide abstract/bio SPEAKER BIO: Dr. Minlie Huang now is an associate professor, deputy director of the AI Lab. of Dept. of Computer Science and Technology, Tsinghua University. He won MSRA collaborative research award in 2019, and Hanvon Youngth Innovation Award in 2018. He won IJCAI-ECAI 2018 distinguished paper, NLPCC 2015 best paper, and CCL 2018 best demo award, ACL 2019 best demo candidate. His work on Emotional Chatting Machine was reported by MIT Technology Review, the Guardian, NVIDIA, Cankao Xiaoxi, Xinhua News Agency, etc. He has published 60+ papers in premier conferences such as ACL, AAAI, IJCAI, EMNLP, WWW, SIGIR, and highly-impacted journals like ACM TOIS, IEEE TASLP, Bioinformatics, JAMIA etc. He served as area chairs for EMNLP 2019, ACL 2016, EMNLP 2014, EMNLP 2011, and IJCNLP 2017, and Senior PC of IJCAI 2017/IJCAI 2018(Distinguished SPC)/AAAI 2019/IJCAI 2019/AAAI 2020, and reviewers for ACL, EMNLP, NAACL, IJCAI, AAAI, and reviewers for journals such as TOIS, TKDE, TPAMI, etc. He was supported by several NSFC projects and one key NSFC project. ABSTRACT: The controllability of natural language generation is one of the most fundamental problems in modern natural language processing. The current neural-network-based natural language generation model generally has the issue of controllability: such as repetition, incoherence of context, and inconsistency of semantic logic. This talk will focus on the current academic community's attempts at this issue, and introduce the team's preliminary attempts to use word type information, knowledge, planning, etc. for controllable text generation. |
114:45-15:25 |
Diving into the Training for the Sequence-to-Sequence Model
Yang Feng, Associate Professor in ICT/CAS show abstract/bio hide abstract/bio SPEAKER BIO: Yang Feng, Associate Professor in Institute of Computing Technology, Chinese Academy of Sciences (ICT/CAS). She was selected into “New Baixing Talent Introduction Program” of ICT/CAS in 2017. Her research and main contributions focus on machine translation and dialogue and her work of “bridging the gap between training and inference for neural machine translation” has won Best Long Paper Award of ACL 2019, the only winner in China. She has published over 40 papers in the top-tier conferences for natural language processing such as ACL, EMNLP, COLING, NAACL and so on. Her work on system combination and low-resource language translation has won the first several times in the authoritative evaluation competitions of machine translation including NIST, IWSLT and CWMT. She has served as Area Co-chair for COLING 2018, Student Forum Chair for CCL 2018-2019 and Forum Chair for CCMT 2019. She is presiding over projects supported by National Key R&D Program of China and National Natural Science Foundation of China. ABSTRACT: Natural language generation is mainly based on the sequence-to-sequence model which usually adopts the teacher forcing strategy for training. Teacher forcing works in the way of comparing the generated translation with the ground truth word by word and minimizing the cross-entropy between each word pair. However, the mechanism is not total reasonable and the unreasonable aspects make it suffer from discrepancy between training and inference, being subjected to word order, lack of sequential information, wrong punishment for diverse translation and so on. This talk focuses on these problems of teacher forcing and discusses the solution to them. |
| 15:25-16:00 |
Break |
16:00-16:30 |
Sentimental Text Generation
Xiaojun Wan, Professor in Peking University show abstract/bio hide abstract/bio SPEAKER BIO: Xiaojun Wan is a professor in Institute of Computer Science and Technology of Peking University and the director of Lab of Language Computing and Web Mining. He obtained his bachelor, master and doctor’s degrees from Peking University. His research interests include document summarization, text generation, sentiment analysis and semantic computing. He has been editorial board member or associate editor for TACL, Computational Linguistics and Natural Language Engineering. He is PC co-chair of EMNLP-IJCNLP 2019 and also served as area chair or SPC for a few international conferences, including ACL, NAACL, EMNLP, IJCAI and AAAI. He has received ACL 2017 Outstanding Paper Award, IJCAI 2018 Distinguished Paper Award, CAAI WUWENJUN Technical Innovation Award (2017) and CCF NLPCC Distinguished Young Scientist (2017). ABSTRACT: Natural language generation has wide applications in many fields. Sentimental text generation is one of the frontier research directions in the field of natural language generation, and it aims to automatically generate natural language texts with specific sentiment attributes, thus meeting the demands of personalized and opinionated text generation in various scenarios, e.g., comment generation and emotional chatting. In this talk, I will introduce the recent advances of sentimental text generation. |
| 16:30-17:00 |
NLG’s Application in the Media Industry
Dengping Zhou, The Founder of Beijing Wribot Intelligent Technology Co. show abstract/bio hide abstract/bio SPEAKER BIO: Graduated from Wuhan University, Dengping Zhou was the general manager of Founder Online Publishing Business Division. Mr. Zhou founded Jiebao Data, a company focuses on news data analysis and was merged by BlueFocus, the largest communication group in Asia, in 2015. In 2019, Mr. Zhou established Beijing Wribot Intelligent Technology Co., Ltd, a company that focuses on the engine development and application of the NLG. Till now, Wribot has created innumerable news contents on self-media platforms and gained more than 100 million views.
ABSTRACT: Media is one of the most important carriers and spreaders of the natural language. In recent years, many “new species” of AI have been born in the media industry. For example, News Writer Robot, Content Editing Robot Assistant, Content Creator & Distributor Robot, AI TV host, Advertising Robot, etc. AI is permeating into our life via the media industry, which is able to affect our vision and mind. This report will present the application and development paths of AI, especially the NLP technology, from perspectives of challenges and chances brought by AI, as well as the typical application of AI in the media industry. |
| 17:00 |
END |
◆ Natural Language Generation and Its Application
Host: Beijing Wribot Intelligent Technology Co., Ltd.
Chair: Jianhua GAO
Description: NLG (Natural-Language Generation) is a science of studying how to generate high-quality readable text information by machine. According to different information inputs, NLG can be classified into “data to text”, “text to text”, “meaning to text”, “image to text”, etc.
In recent years, a great number of cutting-edge NLG researches have been made in the AI field. Moreover, some of these researches have been already applied into the industry. This workshop will introduce latest scientific achievements about NLG and breakthroughs made by NLG in industrial applications. Therefore, the workshop will include talks from leading experts of NLG from both academia and industry.
| Time |
Details |
| 14:00-15:00 |
Lifelong and Continual Learning on the Job
Bing Liu, Professor, University of Illinois at Chicago / Peking University show abstract/bio hide abstract/bio SPEAKER BIO: Bing Liu is a distinguished professor of Computer Science at the University of Illinois at Chicago (UIC). He received his Ph.D. in Artificial Intelligence from University of Edinburgh. Before joining UIC, he was a faculty member at the School of Computing, National University of Singapore. His research interests include lifelong learning, sentiment analysis, chatbot, natural language processing (NLP), data mining, and machine learning. He has published extensively in top conferences and journals. Three of his papers have received Test-of-Time awards: two from SIGKDD (ACM Special Interest Group on Knowledge Discovery and Data Mining), and one from WSDM (ACM International Conference on Web Search and Data Mining). He is also a recipient of ACM SIGKDD Innovation Award, the most prestigious technical award from SIGKDD. He has authored four books: two on sentiment analysis, one on lifelong learning, and one on Web mining. Some of his work has been widely reported in the international press, including a front-page article in the New York Times. On professional services, he has served as program chair of many leading data mining conferences, including KDD, ICDM, CIKM, WSDM, SDM, and PAKDD, as associate editor of many leading journals such as TKDE, TWEB, DMKD and TKDD, and as area chair or senior PC member of numerous NLP, AI, Web, and data mining conferences. Additionally, he served as the Chair of ACM SIGKDD from 2013-2017. He is a Fellow of the ACM, AAAI, and IEEE. ABSTRACT: It has become increasingly apparent that it is difficult to build a truly intelligent agent like a self-driving car or a chatbot that function well in a real-world open environment via pre-training using human labeled data. One reason is that it is impossible to label data to cover all possible real-life scenarios that an agent may encounter. Thus, it is necessary that the agent learns by itself on the job in its interaction with the environment continually, retain the learned knowledge, and use it to learn more and better in the future. When faced with an unfamiliar situation, it can adapt its knowledge to deal with it and learn from it. This general learning capability is one of the hallmarks of the human intelligence. Without it, an AI agent will probably never be truly intelligent. In recent years, we have seen an emerging and growing research trend under the names of lifelong learning, continual learning, open-world learning, etc., that try to endow AI agents with this learning capability. In this talk, I will introduce this topic and discuss some recent research. In this work, we also strive to generate results or actions that are explainable. |
| 15:00-15:30 |
Paper Oral Reports |
15:25-16:00 |
Break |
| 16:00-17:45 |
Paper Oral Reports |
◆ Explainable AI
Host: AI Lab, Lenovo Research
Chair: Yangzhou DU, Gang CHEN
Description: Deep learning has contributed to the recent big progresses in artificial intelligence. In comparison to the traditional machine learning methods such as decision tree, support vector machine, deep learning has achieved significant improvement in predication accuracy. However, deep neural network is very weak in interpretability and explainability of the reasoning process and decision results. In fact, DNN is a blackbox both for developers and for the users. Some people consider DNN and deep learning in the current stage as alchemy, not as science. In many real world applications such as business decision support or optimization, medical decision or diagnostics support, explainability and interpretability and transparency of our AI systems are very important, for the users, for the people who are affected by the AI systems, and in particular, for the researchers and developers. In the recent years, the explainability and explainable AI have received big attention from research to industry. This workshop will provide a forum for sharing the insights about the scope and research topics about explainable AI, exchanging the state-of-the art of explainable AI research and applications, discussing future direction and further work in this area.
