Hi, This is Hao 👋

I work on agentic systems and foundation models, while exploring interpretable methods like Causal ML, state reconstruction and Physics Informed ML.

Previously, I helped building an AI startup in San Jose–Paradigm.Study. My work LLMs — a decomposition story and some following up LLM-dynamics research ideas has been applied to benchmarking OpenClaw and PRs used by their official benchmark repo.

My Google scholar page, most focus on ML+Bio and interpretations of large models. I like to throw information theory and other applied maths, stats into the works (although I am not always lucky). If you really have nothing to do, read my papers.

But the field that interests me more is on AIs that learns from structures from sciences and informs new structures for sciences. Following FuhaiLiAILab I’ve been exploring AI+bio. One of our work is the OmniCellAgent which helps bring specialized data/tools and literature search together to help cure disease link, experimental we also build more interpretable and futuristic foundation models with causal and physics structures…

My history? I used to compete in model airplane aerobatics contests. I learned about chaos and fractals (the picture above) because of aero and multi-body problems I knew of when designing model aircrafts. I took math as undergrad at UC Irvine, but wanted to build something, so I took a few hackathons (and I won some mini projectors😄), then went to study Statistics in Columbia University. Why I chose to do research? I noticed there are fundamental gaps, oversights and alternative theories and techniques that has not been explored, and throughout my experiences there are many surprises and beautiful findings yet to be looked at–especially from the problems that are jointly given to us by the nature and the logics.

I embrace and criticize AI at the same time🤔–but what can I do? If we can bring a little bit of light, then do it. 🔭

Cheers.

Hao Li

One of the Fractal images I rendered late 2019

Figure from my site for fractal rendering: Gallery in Fractal Dimensions https://galleryinfractaldimension-temp.weebly.com/

(Earlier) Project Showcase

Machine Learning

Earlier projects on Kaggle and on Hackathons

sk_in_torch: Using PyTorch to fit scikit-learn models–and using GPU/CPU. (Python)
ML_BioStats: Biostatistics problems on open datasets, Kaggle, hackathons, and eval.ai. (Jupyter Notebook)
ChatMyArchive: A lightweight local file source searching repo for multiple files. For 2024 Brown Hackathon. (Jupyter Notebook)
LinuxMLTroubleshoot: Problems encountered in building machine, trying to remove windows, install linux, conda, pytorch with cuda, rstudio, devtools and more–and how I got through. (Shell)

Math Modeling Code and Math Visualizations

Previewer: Previewing datasets with save and systematic statistical approach. (R) submitted to2021 Shiny R Contest try it

Sample code for Meerschaert’s Mathematical Modeling (4th ed.), special thanks to the suggestion from the author of the book

R sample codes for Meerschaert's Mathematical Modeling (4th ed.): (R)
Sage math sample codes (SageMath, Python 2.7 jupyter notebook or use CoCalc)
rMathModeling: General Math Modeling Tools for R. (R)

Fractals

HypercomplexFractal: Hypercomplex fractal rendering with R. (R)
Gallery in Fractal Dimensions: Figure from my site for fractal rendering.
bryanhanson/LindenmayeR: Helped with visualizing Lindenmayer or L-Systems in R. (R)
1d-mapping-explorations: A project to utilize my previous R package (not documented yet) SeriesXplorer on 1d mappings. Detection of periodicity, chaotic behaviors and animated visualizations are the major tasks. (R)
Interactive nb files with mathematica: For illustration of math concepts, with interactive slidebars. (Mathematica)

Model Aircraft Design and Computing

ModelPlanePower: (R)
eXpand: Code snippet for expanding 3D surfaces to a 2D projection - useful for building models of complicated 3D forms by folding 2D papers/foamboards or cardboards. (Julia)
rAviExp: R library for Model Aviation Exploratory design. (R) In deed you can still use some web UIs here Least climbing speed bound

design a plane with conventional concept

lift/drag and ratio and optimal Cl/Cd w.r.t. angle of attack