// research highlights
Physics-informed search algorithms that identify promising new materials before they're ever synthesized.
MCTS · inverse design
Integrating machine-learned reward and acceleration methods into MCTS for the inverse design of heavy-fermion (superconducting) quantum materials. Developed with Los Alamos National Laboratory.
Read more →Building faster, more accurate, and more data-efficient interatomic potentials for molecular dynamics simulation.
data curation
Group-aware adaptive pruning of ML-IAP training data using QUESTS entropy metrics, reducing test force-MAE by up to 37% versus a single global bandwidth.
Read more →
Bayesian optimization
A multi-level optimization algorithm that learns cluster-based training data weights for bespoke interatomic potentials, improving force-constant and RDF prediction by an order of magnitude over uniformly-weighted training.
Read more →
simulation efficiency
Multi-layered machine-learned interatomic models for enhanced simulation efficiency, built on the ChIMES framework.
Quantifying what a model does and doesn't know by analyzing the structure of its training and simulation data.
fingerprinting
A quantitative framework for analyzing ML-IAP training and simulation data via graph-based cluster fingerprints, enabling novelty detection and dissimilarity scoring for active learning and uncertainty quantification.
Read more →
Riemannian geometry
Used Riemannian geometry to analyze symmetric positive-definite matrices (covariances, Hessians, images) for anomaly detection in chemical process monitoring, with Dow Chemical and Argonne National Laboratory.
Read more →