PhD CandidateSchool of Information
University of Michigan
Office: 105 S State St, North Quad 4385, Ann Arbor, MI 48109
See My Name in Chinese
I am a doctoral candidate at the University of Michigan, School of Information. This personal webpage is a bit atypical, because it is full of my own story:) If you prefer not to read it, here is my CV.
Research KeywordsHuman, Language, Intelligence, Data, Interface
Research DirectionsMy research stays in the center of human-computer interaction, artificial intelligence, natural language processing. In specific, my current research focuses on two areas: (1) how the latest advancement of deep learning technology can deliver new user experience, especially in helping humans deal with complexities in natural language; and (2) what interactive visualization designs can whiten the black box of deep learning models.
Overall, I enjoy thinking about design and technical solutions to the above problems. My diverse backgrounds in computational methods, data analysis, visualization, design thinking, and software engineering allows me to produce research outcomes in various modalities: papers, new datasets, interactive visualization systems, programming IDEs, etc. I hope that these will benefit people in multiple fields. The ultimate goal is to drive technological innovations closer to real human need.
My awesome advisor is Prof. Eytan Adar, who always drives me to think about "why" instead of just "how". I have also really enjoyed working with Dr. Paul Bennett and Dr. Adam Fourney from the then Microsoft Research; Dr. Tom Dean, Dr. Sudeep Gandhe, Johnny Chen, and Dr. Dong Xin from Google; Ningyu Chen and Dr. Jin Yan from Baidu; as well as my former advisor, Prof. Qiaozhu Mei. My bachelor degree was in automation from Tsinghua University, China.
Research HighlightsI have built CodeMend, a system that leverages Stack Overflow posts and Web documents to suggest code modifications based on natural language (see UIST '16 paper). My other projects include mining scholarly articles to predict technology adoption by researchers (see CIKM '13 paper) and mining Wikipedia to support topic exploration via entity set expansion (see WSDM '16 paper).
I am also interested in improving the interpretability of machine learning models, in particular, neural network models. I have built visualization systems that make understanding and debugging neural language models easier (see WEVI and LAMVI). I have also written a tutorial on word2vec, which a lot of people found useful (word2vec is a popular package that learns word embedding vectors from large-scale text corpora by people then in Google.).
PublicationsGoogle Scholar ResearchGate
Workshop Papers and Technical Notes
TeachingA lot of students have praised me for being a committed and patient student instructor that cares to details, but not all of them turned in their homework on time. Nonetheless I truly enjoyed working with these students, and I believe teaching is one of the most enjoyable/rewarding experience one can have. I have taught the following courses as a teaching aid to Prof. Eytan Adar and Prof. Chuck Severance:
AwardsI have received the best workshop paper award from ICML visualization for deep learning in 2016.
I received multiple awards for academic excellence from Tsinghua University at Beijing, back in the times when exam scores (which I used to be good at) are the dominant metrics for such rewards. I believe now they are better.
Fun FactI am very enthusiastic about general aviation. I am a certified private pilot in the United States, currently affiliated with Michigan Flyers. I love taking friends up for sightseeing flights. I have flown a number of Detroit River tours, San Francisco Bay tours, and Pudget Sound tours. Contact me if you are around and are interested for a fun flight!
Perhaps one day I will work on using text mining + HCI methods to enhance aviation safety (e.g., finding opportunities to improve the aircraft troubleshooting instructions or help the understanding of the FAA publications which are extremely important but no one can afford the time to read through).