Chao Zhang @ LAMDA, NJU-AI

Chao Zhang (C. Zhang)

Ph.D. Student,   LAMDA Group
School of Artificial Intelligence
State Key Laboratory for Novel Software Technology
Nanjing University (Xianlin Campus)

Supervisor: Prof. Yang Yu



Since September 2021, I became a Ph.D. candidate of School of Artificial Intelligence in Nanjing University and a member of LAMDA Group, which is led by professor Zhi-Hua Zhou.  Before that, I received my M.Sc. degree in June 2019 from LAMDA Group also under the supervision of Prof. Yang Yu.

I am an AI researcher with well-balanced academia and industry experience. In my previous work, I deeply participated in practical projects related to vision, robotics and data mining, and successfully landed. At present, I pay more attention to how to implement reinforcement learning into real business to release the power of intelligent decision-making.   


Research Interest

Reinforcement learning has been shown to surpass human decision-making ability in the game, and the main reason behind this is that the game scene is a natural same-training-application environment for online learning. However, such a natural environment often does not exist in real business scenarios. Therefore, how to effectively combine the advantages of decision-making ability in reinforcement learning with the excellent ability of data mining in machine learning to improve the efficiency of real-world tasks under the fully offline setting is my main research direction. Relevant sub-areas include:

  • Environment modeling combined with expert experience or physical facts
  • Offline reinforcement learning with safety constraints
  • Good initial decision-making performance for cold start or scenarios with few recorded data


Publication & Patent

  • Chao Zhang, Yang Yu, Zhi-Hua Zhou. Learning environmental calibration actions for policy self-evolution. In: Proceedings of the 27th International Joint Conference on Artificial Intelligence (IJCAI'18), Stockholm, Sweden. (PDF) (Poster) (Github)
  • 俞扬,张超,周志华. 一种基于模拟器训练的机器人控制方法. 201810054083.0, 2018.01.19.
  • 张超,徐易楠,杨振宇. 基于强化学习的关键词抽取方法. 201810774634.0, 2018.07.13.
  • 秦熔均,刘泽琳,张超. 一种利用模拟环境重构投融资行为的方法. 202110311327.0, 2021.03.24.


1、Polixir, 2020.12-2021.09

Business Leader

Business Scope: Automotive Manufacturing, Thermal Power Energy, Intelligent marketing.

2、Huawei, Car BU, 2019.07-2020.12

Algorithm Engineer (FX人才计划), Research on visual perception of autonomous driving

Skills: Detection and Tracking, Semantic Segmentation, Point Cloud Perception, 2D-3D Transformation

3、ZhuiYi technology (Internship), 2018.03-2018.07

Algorithm Engineer, Research on Chatbot

My main work: Extracting Context Key Words

4、Tencent AI Lab (Internship), 2017.11-2018.01

Algorithm Engineer, Research on Deep Reinforcement Learning

My main work: Building the virtual environment, training the control policy and adapt to the phsical robot


Academic Activities

  • Giving a Spotlight talk on policy self-evolution in MLA, 2018
  • Attending the 35th International Conference on Machine Learning (ICML). Stockholm, Sweden, July, 2018
  • Attending the 27th International Joint Conference on Artificial Intelligence (IJCAI). Stockholm, Sweden, July, 2018
  • Attending Vision And Learning SEminar (VALSE). Xiamen, April, 2017
  • 张超,胡琳梅,田天,唐杰,俞扬. 由学术展开交流,从应用探讨发展——记AAAI 2017. 中国计算机学会通讯, 2017, 13(3):72-75. (PDF)
  • Attending the 31st Association for the Advancement of Artificial Intelligence (AAAI). San Francisco, USA, February, 2017
  • Attending the 14th China Workshop on Machine Learning and Applications (MLA). Nanjing, November, 2016, 2018
  • Attending China National Computer Congress (CNCC). October, 2016, 2017, 2018

Awards & Honors

  • 华为FX顶尖人才计划,2019
  • 南京大学优秀毕业生, 2019
  • 南京大学硕士国家奖学金 , 2018
  • 南京大学优秀研究生,2018
  • 腾讯犀牛鸟首届精英研究生计划,2017



Address: Chao Zhang, National Key Laboratory for Novel Software Technology, Nanjing University, Xianlin Campus Mailbox 603, 163 Xianlin Avenue, Qixia District, Nanjing 210023, China
(南京市栖霞区仙林大道163号, 南京大学仙林校区603信箱, 软件新技术国家重点实验室, 210023.)

Last modified: 2021-09-15
The end