Joy Jingyi Xu

About Me

I am an advanced software engineer at Siemens NX CAD (Computer-Aided Design). I worked in NX Diagramming team.

Before that, I was a doctoral student at Northeastern University's Network Science Institute, advised by Tina Eliassi-Rad. My research at Northeastern focused on comparing graph embedding methods, the interaction of math, computer science and network science. I was especially interested in combining machine learning methods with traditional algorithms and in developing models for estimation and prediction. My broader research interests included network science, machine learning, data mining, and algorithms. I left my PhD program to focus on building software, learning beyond formal boundaries, exploring and pursuing faster feedback loops through real-world products.

I earned my master's degree in Computer Science from the University of California, Los Angeles (UCLA), in 2018, with a focus on machine learning. I published my research with my labmates, and professor Guy Van den Broeck in ICML 2018.

I earned my bachelor's degree in Mathematics, with a minor in Computer Science, from the University of Michigan in 2014. Before returning to graduate school in UCLA Computer Science in 2016, I worked as an actuary in the Asset Liability Management Department at Sun Life Financial in the Boston metropolitan area.

Self-Initiated Project

Peter Lynch Stock Analysis Tool: I built a Python web app that analyzes stocks using Peter Lynch's investment methodology. Enter any stock ticker and get an instant Lynch-style analysis — PEG ratio, 10-bagger potential, dividend-adjusted Lynch formula, debt assessment, and an overall Lynch Score with Buy/Hold/Avoid verdict. In addition, I also made a web browser for visualization.

Software Construction Project: convert research PDFs into audio files so that users can listen to research papers. Current text-to-speech tools also convert table contents, headers, and footnotes to audio, which makes research papers difficult to follow. Our tool identifies and removes URLs, headers, footnotes, and math formulas, improving users' comprehension of research papers. Future work includes making automatic text-to-speech tools read math formulas more naturally. Software Construction Project Joint work with Hanjun Li, Sherry Lin, Andres Abdo, Yunsheng Bai.

Research and Publications

Machine Learning Research: applying constraints on neural networks. Jingyi Xu, Zilu Zhang, Tal Friedman, Yitao Liang and Guy Van den Broeck. A Semantic Loss Function for Deep Learning with Symbolic Knowledge, In Proceedings of the 35th International Conference on Machine Learning (ICML), 2018. 📄 758 citations by 2026

Graph Embedding Research: In an attempt to develop a theoretical understanding of the math behind "Why neural network works so well", I compared 6 network-embedding methods from three broad categories: matrix-factorization, random walk approaches, and deep architecture approaches. We demonstrated that matrix factorization methodscorrelate with deep neural network embedding methods. Understanding Graph Embedding Methods and Their Applications

Algorithm Research: finding an optimal solution for the mutually exclusive set cover problem. (Individual Project) project repository

Robotics Research: evacuation planning using HoloLens and RTAB-Map SLAM. (Joint project with Hangxin Liu, Zhenliang Zhang, and Yixin Zhu) project repository

Course Name	Organization	Year
Robotics: Computational Motion Planning	Coursera	2018
Machine Learning	Coursera Andrew Ng	2017
Algorithms	University of Hawaii	2016
Linear Algebra	MIT OCW	2018

Course Name

Organization

Year

Robotics: Computational Motion Planning

Coursera

2018

Machine Learning

Coursera Andrew Ng

2017

Algorithms

University of Hawaii

2016

Linear Algebra

MIT OCW

2018

Joy Jingyi Xu

About Me

Self-Initiated Project

Research and Publications

Online Classes Taken Outside School