HAO XINHao Xin
Senior Machine Learning Scientist @ Apple
// teaching machines to see, think, and occasionally hallucinate
where mathematics meets imagination
< ABOUT.ME />About
$ cat self_description.txt
Hey there! I'm Hao -- a PhD-wielding research scientist who somehow convinced Apple to let me build AI systems for a living. My day job involves making diffusion models behave, wrangling large language models into being useful, and pretending I understand why transformers work so well.
I went from studying pure math and statistics (because I thought it would be "fun") to training billion-parameter models on clusters of GPUs that cost more per hour than my first car. No regrets.
$ cat background.json
{
"origin": "Nankai University — pure math",
"path": "Math → Stats → ML → Deep Learning → GenAI",
"thesis": "Score-based diffusion for image regression",
"current": "Generative AI for health at Apple"
}
$ _
I've always been drawn to the place where structure meets surprise — the moment a mathematical proof clicks, or a model generates something unexpectedly beautiful.
My path wandered from pure mathematics at Nankai University through the probabilistic landscapes of Wisconsin and Purdue, eventually arriving at Apple, where I now work on generative AI for health — teaching models to create, to understand, and sometimes to see what we cannot.
I believe the best science feels like art: an elegant equation, a well-designed experiment, a diffusion process that turns noise into meaning.
< THE.LORE />Journey
// the commits that made me
the chapters that shaped me
Senior ML Scientist
Apple Inc. // AI for Health Dec 2024 – Present- Generative AI for health — diffusion models, multimodal systems, the works
- Post-training and alignment of LLMs; making them a little less chaotic
- Data pipelines and evals — the unglamorous half that makes the glamorous half work
Senior Data Scientist
Walmart Global Tech // CV & GenAI Jul 2022 – Dec 2024- Computer vision and 3D perception for retail-scale environments
- GenAI pipelines for synthetic data, content creation, and LLM-powered search
- ML inference optimization for production — because accuracy means nothing if it's too slow
Ph.D. in Statistics
Purdue University 2017 – 2022- Thesis on score-based diffusion models for image-to-image regression
- Five years of stochastic processes, probability theory, and learning to love LaTeX
- Also: M.S. in Statistics @ UW-Madison, B.S. in Math & Stats @ Nankai University
< RESEARCH.DB />Research
// focus: score-based diffusion models for 3D regression tasks
exploring how noise becomes signal
Generative Image-to-Image Regression Based on Score Matching Models
Purdue University Graduate School -- Ph.D. Thesis
< BLOG.LATEST />Writing
< CONTACT.IO />Connect
$ echo "Let's connect!"
Whether you want to talk about diffusion models, debate whether attention is really all you need, or just say hi -- my inbox is open.
$ cat contact_info.txt
$ subscribe --newsletter
Get new posts straight to your inbox — no spam, just ML thoughts.
I'd love to hear from you — whether it's about research, a shared curiosity, or just to say hello. The best conversations often start with a simple note.
If you enjoy thoughtful writing on ML and beyond, leave your email — I'll send you new pieces as they come.