Research & projects
Some problems I kept coming back to.
A mix of fusion physics, fluid dynamics, space science, and the compute needed to make them practical.
Neural-operator surrogates for high-frequency RF wave propagation in tokamak plasmas, focused on highly oscillatory complex fields and phase accuracy.
Physics-Informed Neural Operator surrogate for lower hybrid waves — 281.8× faster inference in rectangular domains, 25.8× in TST-2 circular geometry, within a 498 MB footprint.
FNO surrogate for mold-flow analysis in Moldex3D — predicting pressure (MAE < 3 Pa) and melt front time (MAE < 0.3 s).
Regression-based Physics-Informed Neural Networks combining empirical models with deep learning — ~30% RMSE reduction vs. Shue et al. [1998].
Multicore benchmarking of a volatility-momentum equity strategy on an HPC cluster.