Zekang Li (李泽康) Master student NLP Group, Key Laboratory of Intelligent Information Processing, ICT, CAS Email: zekangli97 [at] gmail.com or lizekang19g [at] ict.ac.cn Stay hungry, stay foolish.

Biography

My research interests lie within Natural Language Processing, particularly in Dialogue System and Vision-Language tasks. I am also involved in Cognitive Science and its application in Dialogue System.

News

• [2021/05] Two papers about dialogue pre-training and dialogue consistency evaluation were accepted to ACL2021.
• [2021/04] We will organize Track1: MOD:Internet Meme Incorporated Open-domain Dialog at DSTC10. Welcome to join the challenge!
• [2021/02] I was invited to give a talk in the NSF Future Directions Workshop on the Automatic Evaluation of Open-Domain Dialog (AED).
• [2021/02] One paper about Audio-Visual Scene-aware Dialogue was accepted to IEEE/ACM TASLP.
• [2021/02] Our project (Science Fiction AI-Human Co-writing) won the First Prize in the 6th National Youth Artificial Intelligence Innovation, CAAI, advised by Yonggang Wang and Ming Zhou.
• [2020/11] One paper about Interactive Dialogue System was accepted to DSTC9@AAAI2021.
• [2020/10] We participated in the DSTC9 Interactive Dialogue Evaluation Track. We tied the 1st in sub-task1 and got the 3rd place in sub-task2.
• [2020/08] Our project (Science Fiction AI-Human Co-writing) won the 1st place in the Innovation Track at Deecamp 2020.
• [2020/04] One paper about Dialogue State Tracking was accepted to ACL2020.
• [2019/11] One paper about Audio-Visual Scene-aware Dialogue was accepted to DSTC8@AAAI2020.
• [2019/10] We won the 1st place in the DSTC8 AVSD Track.
• [2019/05] One paper about Document Grounded Conversations was accepted to ACL2019.
• [2018/11] I join WeChat AI as Research Intern.

Education

• Huazhong University of Science and Technology, Wuhan, China

  Sep. 2015 – Jun. 2019
  B.E. in Computer Science and Technology
  
• Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China

  Sep. 2019 – Jun. 2022 (expected)
  M.S. in Computer Science and Technology
  

Publications (Selected)

• Conversations Are Not Flat: Modeling the Intrinsic Information Flow between Dialogue Utterances
  Zekang Li, Jinchao Zhang, Zhengcong Fei, Yang Feng, Jie Zhou.
  To appear at the Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics (ACL), 2021. (Online)
  
• Addressing Inquiries about History: An Efficient and Practical Framework for Evaluating Open-domain Chatbot Consistency
  Zekang Li, Jinchao Zhang, Zhengcong Fei, Yang Feng, Jie Zhou.
  To appear at the Findings of the Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics (Findings of ACL), 2021. (Online)
  
• Bridging Text and Video: A Universal Multimodal Transformer for Video-Audio Scene-Aware Dialog
  Zekang Li, Zongjia Li, Jinchao Zhang, Yang Feng, Jie Zhou.
  IEEE/ACM Transactions on Audio, Speech and Language Processing (TASLP), 2021.
  

  [Paper] [Bib]

• WeChat AI's Submission for DSTC9 Interactive Dialogue Evaluation Track
  Zekang Li, Zongjia Li, Jinchao Zhang, Yang Feng, Jie Zhou.
  Proceedings of the 9th Dialog System Technology Challenge Workshop in AAAI2021. (DSTC9@AAAI2021), 2021. (Oral)
  

  [Paper] [Bib]

• A Contextual Hierarchical Attention Network with Adaptive Objective for Dialogue State Tracking
  Yong Shan, Zekang Li, Jinchao Zhang, Fandong Meng, Yang Feng, Cheng Niu, Jie Zhou.
  Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL), 2020. (Online)
  

  [Paper] [Bib]

• Incremental Transformer with Deliberation Decoder for Document Grounded Conversations
  Zekang Li, Cheng Niu, Fandong Meng, Yang Feng, Qian Li, Jie Zhou.
  Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL), 2019. (Oral)
  

  [Paper] [Bib]

• An end-to-end automatic cloud database tuning system using deep reinforcement learning
  Ji Zhang, Yu Liu, Ke Zhou, Guoliang Li, Zhili Xiao, Bin Cheng, Jiashu Xing, Yangtao Wang, Tianheng Cheng, Li Liu, Minwei Ran, Zekang Li.
  Proceedings of the 2019 International Conference on Management of Data (SIGMOD), 2019.
  

  [Paper] [Bib]

Manuscripts

• Towards Multimodal Response Generation with Exemplar Augmentation and Curriculum Optimization
  Zeyang Lei*, Zekang Li*, Jinchao Zhang, Fandong Meng, Yang Feng, Yujiu Yang, Cheng Niu, Jie Zhou.
  arXiv preprint arXiv:2004.12429 (2020).

  [Paper] [Bib]

Experience

• WeChat AI, Beijing, China
  
  Research Intern, advised by Jinchao Zhang, Fandong Meng, and Cheng Niu.

  Nov. 2018 – Now
  • Interactive Dialogue: Interactive Dialogue System (DSTC9), Interactive Dialogue Consistency Evaluation (ACL2021), Dialogue Pre-training and Evaluation (ACL2021).
  • Grounded Dialogue Generation: Document Grounded Conversations (ACL2019), Audio-Visual Scene-aware Dialog (TASLP), Meme Dialogue (ACL2021).
  • Dialogue Coherence and Diversity: WAE and examplar based dialogue generation.
  • Task-oriented Dialogue: Dialogue State Tracking (ACL2020).
• Dian Group, Wuhan, Hubei province, China
  
  Group Leader & Founder, AI Group, advised by Xinggang Wang.

  Feb. 2018 – Jun. 2019
  • Management: Founder of AI Group in Dian Group. I expanded the group from 4 members to 20+ members. AI Group got the Best Group award in Dian Group for consecutive two years.
  • Development: We cooperated with famous companies on commercial or research projects, such as
    • 1) AIBot: An Intelligent Customer Service Robot (Beibei Group).
    • 2) AI in 5G: Using Deep Reinforcement Learning to optimize Network Access Control in 5G. (Intel Inc).
    • 3) Real­time Face Detection & Alignment Project.
    I mainly leaded the AIBot project and guided the other projects.
• Wuhan National Laboratory for Optoelectronics, HUST, Wuhan, Hubei province, China
  
  Research Assistant, advised by Prof. Ke Zhou, cooperate with Tencent Cloud.

  Sep. 2017 – Aug. 2018
  • Cloud Database Autotune: Tune tencent cloud database automatically using deep reinforcement learning. Our system significantly outperforms the state-of-the-art tuning tools and the DBA experts. Our paper is accepted by SIGMOD 2019.
  • My work: 1. Analyze cloud database knobs to find which knobs affect performance a lot. 2. Propose the idea to use deep reinforcement learning to tune cloud database automatically, build the tuning system, and design rewards.

Projects (Selected)

• The First Chinese Science Fiction Human-AI Co-writing Experiment
  
  Team Leader, cooperate with Xiangru, Zhengcong, Xiaohao, advised by Yonggang Wang.

  Aug. 2020 – Dec. 2020
  • Project Introduction: This project originated in the Deecamp 2020, Sinovation Ventures, which was lectured by 12 world-renowned AI experts including Andrew NG, Zhihua ZHOU and Kai-fu LEE. We designed and built a Human-AI Co-writing System with the Chinese large-scale pre-trained language model and the large-scale science fiction corpus. We also adapted different writing styles for different writers.
  • Operation: 11 human science fiction writers (including Qiufan Chen, Shi Gu, etc) and AI collaborated to create multiple science fiction stories on topics such as environmental protection, human-computer relationship, gender, and cultural diversity.
  • Social Influence: The People’s Daily, Science and Technology Daily, Kaifu Li, Yahoo News.
• AIBot: An Intelligent Customer Service Robot
  
  Team Leader, cooperate with Beibei Group.

  Mar. 2018 – Nov. 2018
  • Project Introduction: Beibei Group is the biggest e-commerce company for mothers and babies in China. I lead a team of 12 undergraduate students to develop a stable, learnable and reliable intelligent customer service system to replace the original commercial system. The response accuracy and customer’s satisfaction of our system outperforms the original commercial system (Yibot). Now AIbot serves hundreds of millions customers.
  • Methods:
    • Models: We design text classification models (TextCNN, TextLSTM) to select the most proper responses according to users' questions.
    • Active Learning: To get continuous improvement after the deployment, we train a discriminator to identify the bad/low-confidence cases and annotate them by experts. Annotated data will replace the bad/low-confidence cases in the training process.
    • Offline Training and Online Serving: We used X-learning distributed training platform to update our model everyday and used Kafka to meet high concurrency requirements for online serving.
  • Achievement: AIBot achieved the similar accuracy on the response selection with the commercial system Yibot. After one month's tuning, AIBot got 1.7% improvement on the user satisfaction and 2.3% decrease on the turnover rate.

Competitions

• First Prize in the 6th National Youth Artificial Intelligence Innovation, CAAI.

  Feb. 2021
• First Place in sub-task1 and 3rd place in sub-task2 of the IDE Track at DSTC9 Challenge.

  Oct. 2020
• First Prize in 2020 Artificial Intelligence Application Innovation Competition, Huawei Cloud

  Aug. 2020
• First Place of Innovation Track, Deecamp 2020, Sinovation Ventures

  Aug. 2020
• First Place of Audio-Visual Scene-Aware Dialog Track at the DSTC8 Challenge.

  Oct. 2019
• Meritorious Winner in MCM 2018

  Apr. 2018
• Second Prize in CCF Big Data & Computing Intelligence Contest (top 36 / 14988)

  Dec. 2017
• Second Place in 1st HackxFdu Hackathon (2/80)

  Oct. 2016

Honors & Awards

• LvDe Scholarship (one of the top awards in ICT, CAS)

  Jan. 2021
• First-class Scholarship of ICT, CAS

  Oct. 2020
• Outstanding Graduate award (top 3%), HUST

  Jun. 2019
• Outstanding Graduate Thesis award (top 3%), HUST

  Jun. 2019
• Third Prize in 2018 Tencent Rhino-Bird Elite Training Program

  Oct. 2019

Professional Services

• ACL: Reviewer (2021, 2020)
• EMNLP: Reviewer (2021)
• AAAI: Reviewer (2021)