About
High-school student in Taipei working on random matrix theory, on-device machine learning, and the awkward systems work in between — getting things to run where people actually are rather than where they are easy to build.
Hello, I'm Yu, Yao-Hsing 尤耀星, a student at Taipei Municipal Chien Kuo High School. I work on the limiting spectral distributions of random matrices — and then on getting that mathematics to actually run somewhere: a denoiser on a Mac, a full Python runtime on an iPhone, a typhoon track model in a browser tab.
The through-line is stubbornness about the gap between a result and a working thing. A proof about eigenvalue supports became an image denoiser and a PyPI package; wanting PyTorch on an iPad became the first public native PyTorch build for iOS.
Publication
Yao-Hsing Yu. Cambridge Journal for Junior Scientists, Volume 3 (2026), Number 2, pp. 365–382. Published 30 July 2026 by International Press of Boston. Open access
doi:10.4310/CJJS.260626162006
The paper grew out of “Geometric analysis of the eigenvalue range of the generalized covariance matrix”, which took the Bronze Medal in Mathematics at the 2025 S.-T. Yau High School Science Award finals. Advised by Shao-Hsuan Wang (National Central University) and Yang-Kai Lue (National Yang Ming Chiao Tung University).
Education
Currently diving into advanced computer science, applied mathematics, and data analytics. This academic environment inspires me to push the boundaries of what's possible and constantly challenge myself.
Skills & expertise
01
Random Matrix Theory, Stieltjes Transform, M-P Law, spectral distributions
02
PyTorch, HuggingFace transformers, on-device LLMs, scikit-learn
03
Swift, SwiftUI, iOS / iPadOS / macOS apps, HealthKit, CarPlay, Metal
04
Native iOS arm64 builds, Rust (PyO3), C / C++ / Fortran, Metal GPU bridges
05
Windows Store apps, GUI applications, Python packaging, web development
06
Data analysis, Manim animations, matplotlib / Plotly, FFmpeg / Cairo
Interests
I'm fascinated by the convergence of machine learning, mathematics, and coding — especially pushing heavy compute (PyTorch, transformers, LaTeX, Manim) onto devices people actually carry. I spend my time researching random matrix theory, building offline-first apps, and cross-compiling "impossible" libraries to iOS.
Experience & timeline
July 2026
“Limiting spectral distributions of random matrices”, Volume 3 (2026), Number 2, pp. 365–382, open access via International Press of Boston. The peer-reviewed form of the award work from the year before. Read it →
2026
A typhoon-track model running as a quantised ONNX file small enough to sit in a browser tab, driving both the live JMA overlay and hindcasts of past storms on the Typhoon Tracks page. The interesting part was the honesty problem: knowing, and saying, how much of a drawn track is the model and how much is persistence.
2026
App Store & Mac App Store — ManimStudio, EigenDenoise, SidecarBridge, WhisperKit, GPS-location-app.
Microsoft Store — ManimStudio, Generalized Covariance Matrix, t-SNE Visualization.
EigenDenoise and Generalized Covariance Matrix are the paper made usable: the eigenvalue-support result turned into an image denoiser. Real download numbers for all of them are on the Store Stats page — including the ones that barely registered, since publishing only the flattering numbers would defeat the point.
2026
python-ios-lib — a Python 3.14 runtime for iOS and iPadOS with 30+ native libraries, including the first public native PyTorch build on iOS and a Metal GPU bridge worth a 2–10× on-device training speedup.
CodeBench — an offline developer and scientific workstation for iPad and Mac: Monaco editor, a real terminal, C/C++/Fortran, on-device pdflatex, and local LLMs.
Bootbox and the WebAssembly work (qemu-wasm, container2wasm) — booting whole operating systems in a browser tab.
2026
rmt-denoise (random-matrix image denoising, the library counterpart to EigenDenoise), cairometal (pycairo-compatible 2D vector graphics on the Apple GPU via Metal — the first GPU cairo on PyPI), python-to-binary (Python → machine code with no C toolchain, assembler, linker, or dependencies at all), narrate, and ollama-installer. Download numbers — with mirror traffic separated out — are on the PyPI Stats page.
2025
2025 丘成桐中學科學獎總決賽
Geometric analysis of the eigenvalue range of the generalized covariance matrix — explicit support bounds for two-point limiting measures, via the Intermediate Value Theorem and Stieltjes transform inversion.
Bronze in Mathematics at the Grand Finals, after Silver at the Asia regional. Applications in statistics, signal processing, and wireless communications. Published in peer-reviewed form the following July.
2023 – 2024
Beginning the random matrix theory work at Chien Kuo High School, alongside early AI projects — and putting the first applications on the Microsoft Store, which is where the habit of measuring what people actually download started.
2021 – 2023
Projects in mathematical modelling and spectral distribution analysis — the groundwork the later papers rest on.
2019 – 2021
Working through programming languages and tools on personal projects, with nobody assigning any of it.