Developing AI models to support scientific discovery
I’m a PhD student at the University of Michigan developing foundation models for scientific discovery. My research spans theory, public health, and precision medicine, with a focus on enabling interpretable and generalizable predictions.
My work includes:
- (Simulation-Grounded Neural Networks) SGNNs: A general framework that uses mechanistic simulations as synthetic supervision to train neural networks for scientific tasks. SGNNs enable forecasting, classification, and inference even when real-world data is sparse, noisy, or missing. SGNNs unify the interpretability of mechanistic models with the flexibility of deep learning, and introduce back-to-simulation attribution, a new form of mechanistic interpretability.
- Mantis: A foundation model for infectious disease forecasting that outperforms baselines across 35+ evaluations.
- Optic: A real-time disease surveillance system that combines simulation-pretrained detection models with robust anomaly and seasonality analysis. Optic achieves state-of-the-art outbreak detection accuracy and requires no disease- or region-specific training—enabling immediate deployment in low-resource public health departments without tuning.
I’m currently working on solving some new technical problems with SGNNs and extending their applications into precision oncology.