Chengrui Zhang
I study deep learning for partial differential equations, with earlier work in semi-supervised learning and feature selection. This is where I write about equations, models, and computational experiments.
Outside research, I build open-source tools, maintain my Linux workspace, and explore coding agents in everyday development.
GitHub activity
Now · Oct 2026Studying neural operators for the Monge–Ampère equation, using finite-difference and Newton solvers for numerical baselines and training data.
Interests and tools
- Research
- Programming
- Scientific computing
- Everyday tools
Research
- 2024 – Present
Operator Learning for the Monge–Ampère Equation
M.S. research · Scientific machine learning
Researching neural-operator methods for the fully nonlinear Monge–Ampère equation, using finite-difference and Newton solvers as numerical baselines, and maintaining code to generate synthetic instances with prescribed boundary conditions.
- 2022 – 2023
MESA: Multi-view Feature Selection
Student lead · NTPIE Undergraduate Research Grant
Extended semi-supervised feature selection to multi-view data using tensor low-rank constraints for cross-view consistency. Released the MATLAB implementation and prepared the project report.
Publications
- 2022
Semi-supervised Feature Selection with Soft Label Learning
IEEE/CAA Journal of Automatica Sinica. First author
- DOI
- Code
@article{zhang2022softlabel, title = {Semi-supervised Feature Selection with Soft Label Learning}, author = {Zhang, Chengrui and Zhu, Lei and Shi, Dan and Zheng, Jiecai and Chen, Haibao and Yu, Bo}, journal = {IEEE/CAA Journal of Automatica Sinica}, year = {2022}, doi = {10.1109/JAS.2022.106055}, }
Projects
The source and Astro theme behind this site, with equations, numbered theorems, citations, and sidenotes for writing in Chinese and English.
A freelance order-management tool for agents, built with Rust and SQLite to track orders, deliveries, and payments, with a terminal UI.
My Gentoo workspace configuration, managed with GNU Stow, covering GNOME, Zsh, Neovim, and command-line tools.
Operator learning for PDEs with attention mechanisms, with experiments organized using PyTorch Lightning and Hydra.
MATLAB code for semi-supervised feature selection with soft-label learning.