CV

Education

• Ph.D in Computer Science, University of California, Davis, 2023-Present
• M.S. in Computer Science, University of California, Davis, 2021-2023
• B.S. in Computer Science, Tongji University University, 2016-2020

Work experience

• Microsoft Research: Software Engineer Intern
  • Designed a high-frequency stock trading data detection algorithm for stock price prediction and analysis, where the important turning points of intra-day stock trading can be well captured and visualized
  • Achieved 29.15% improvement on the Information Coefficient (IC) of the stock price forecast task based on a quick response to important intra-day transaction records, which were from some representative active stocks in the trading market
  • The research paper for the project results is under-reviewed
• Zhihu Inc: Software Engineer Intern
  • Crawled and cleaned millions of Q&A text and pic&text pairs from user posts on Zhihu.com, and built up two large-scale datasets for the pre-training models by MapReduce, Spark, Scala, and SQL
  • Designed and implemented the pre-training of Phase-Granularity BERT Language Model, which used a dictionary of 920,000 phrases commonly used in Chinese to shorten the search engine response time
  • Solved GPU’s Out of Memory (OOM) problem caused by model’s large-scale trainable parameters; shortened model’s computing time 31% compared with Character-Granularity BERT model in searching
  • Implemented the Pre-Training of a Multi-Modal (picture-text) model for improving search quality based on Contrastive Language-Image Pre-training, improved the Recall Rate (by 20% on R@1, 30% on R@5 and 30% on R@10) of image and text retrieval by using our model compared to the baseline model Wenlan, a larger Chinese multi-modal pre-training model in terms of size
  • Two patents (CN113326693A, CN113283551B) are issued for the project results
• Shanghai Qiyue technology: Software Engineer Intern
  • Realized a chemistry experiment operation scoring system software based on target detection algorithms
  • Used multi-processing module to detect two perspectives of video frame at the same time and improve the performance of scoring system and the final accuracy of the algorithm is over 90%

Service

• Serve as reviewer for EMNLP 2023
• Serve as reviewer for ARR February May July October 2025