Yongzhe Zhang is a talented research scientist who has effectively integrated machine learning and other technological innovations into his work.
Educational Proceedings
Yongzhe has gained a diverse academic background, having earned the following degrees:
- Bachelor of Science in Mathematics and Applied Mathematics with a National Scholarship Honor from Sun Yat-sen University
- Master’s degree in Mathematics during his enrollment at the University of Wisconsin-Madison
- Ph.D. in Mathematics at the California Institute of Technology.
As an academic leader, Yongzhe earned the National Scholar Prize, Sun Yat-sen University’s first prize, for multiple years because of his outstanding GPA. At UW-Madison, he competed in the Second UW Math Competition, achieving second place out of 17 other intellectual students.
At the California Institute of Technology, Yongzhe had the opportunity to publish several entries about his mathematical research:
- “Superconvexity of the Heat Kernel on Hyperbolic Space”: Published in the Proceedings of the American Mathematical Society journal, this research article proves Bernstein’s conjecture and analogizes Huisken’s formula in mean curvature flow in hyperbolic dimensional space.
- “Nonconvex ancient solutions to curve shortening flow”: Written with the assistance of Connor Nilson, Ilyas Khan, and Sigured Angenent, Yongzhe discovered a solution to the curve shortening flow for all values of negative time. This study was published in the Transactions of the American Mathematical Society journal.
An Era of Algorithm Engineering and Scientific Research
Yongzhe’s exceptional theoretical mathematics background has been critical in his career success. He excels in methodical thinking and intuition to generate ideas and solutions in his current work.
As an algorithm engineer intern at Alibaba Group, Yongzhe spearheaded a solution for taobao.com, China’s largest online platform. He developed a Hyper-GraphSAGE network using PyTorch to predict the ETA of products that the platform sold, thereby reducing the mean absolute error of initial ETAs by 2.6 hours (14.2%) and increasing prediction accuracy from 27% to 35%.
At Meta Platforms, Yongzhe has worked extensively as a research scientist with extremely large machine learning models. Compared to typical models that use about 50 gigabytes of memory, Meta’s models consume roughly 500 gigabytes. His role has involved debugging and refining the robustness of the models, which is complex because of the stochastic nature and size of the models.
Yongzhe effectively used his mathematical and statistical methods to develop the debugging system for large ads machine learning models. He was able to build distributed systems that track, monitor, and debug the operation of recommendation models and provide optimization. For example, he built a data quality monitor for the Ads Training Data Infra team, resulting in a decrease in the false alarm rate from 80% to less than 20%.
Future Objectives
As Yongzhe learns more intricate concepts about the nature of machine learning, he aspires to lead and innovate others. He will guide research scientists from tech companies and universities and encourage junior researchers to achieve their passions.
Yongzhe also plans to focus on advancing the theoretical aspects of machine learning, particularly in understanding the limitations and potentials of current algorithms. He envisions collaborating with interdisciplinary teams to tackle complex problems, ultimately pushing the boundaries of artificial intelligence and data science.