Yang
Deng
Charting the space between physics and machine learning.
I'm an AI Research Scientist at Axiomatic AI, working on verifiable AI for photonics and beyond in science. Before that, I was an early engineer at an AI startup building simulation tools, and did my Ph.D. at Duke on machine learning for metamaterials.
Live demo
Inverse thin-film design, checked by physics.
Sixty-four candidate coatings descend through a neural surrogate toward a reflectance you choose. An exact solver then re-scores every one — the network proposes, physics decides.
- Draw
- Propose
- Verify
64 seeds · 200 steps∇ₓ ‖f̂(x) − R*‖²
What I’m thinking about
Benchmark design
Automated verification for simulation-based physics discovery.
Physics-informed ML
Surrogates, inverse design, active learning — under physical constraints.
Foundation models for science
Transformer-based LLMs as high-dimensional scientific regressors.
Accelerated optimization
Bayesian methods that make expensive simulations cheaper.