About

I'm Guang Yang (杨光). I work at the intersection of Physical AI,multi-agent systems, agent evaluation, andAI for science.

My PhD was in AI for Earth Science, with a focus on fluid simulation, geostatistics, and uncertainty quantification (UQ). That background is not a footnote — it's my edge. In earth science, a prediction without error bars gets desk-rejected. I bring that same discipline to AI agents, where "point estimate as truth" is still the norm.

Education

I earned my Ph.D. at Stanford University, in Energy Resources Engineering (School of Earth, Energy & Environmental Sciences). My doctoral research centered on uncertainty quantification and prediction under uncertainty — applying machine learning to problems like contaminant transport and reservoir production forecasting, where quantifying what you don't know matters as much as the prediction itself.

That training in geostatistics, subsurface flow simulation, and rigorous UQ is the foundation everything I do now is built on. → Read my Stanford thesis.

The through-line

Everything I work on is a projection of one question:

How do we let multiple agents make trustworthy, evaluable decisions under the uncertainty of the real world?

• UQ / geostatistics — my native language for uncertainty.
• Fluid simulation / AI4S — real physical-world modeling, not toy environments.
• Physical AI / MAS — the vehicle for decision and execution.
• Agent evaluation — the "trustworthy & evaluable" part, and the least-solved problem in agents today.

How I work in public

Honest over hyped. I'd rather publish a negative result I understand than a positive one I can't defend. If a method of mine doesn't work, I'll tell you exactly where and why. In 2026, that's a feature, not a bug.

Find me on GitHub · X · LinkedIn.(X / LinkedIn links to be filled in)