Published in arXiv, 2025
This paper argues that epidemic forecasting ensembles often underperform because constituent models share correlated errors. It demonstrates that diversity in modeling approaches, not just individual accuracy, is the key driver of ensemble performance, and proposes strategies to design ensembles that maximize complementary information across models.
Published in arXiv, 2025
The paper presents a foundational theory for Simulation-Grounded Neural Networks (SGNNs), showing that (i) they implement amortized Bayesian inference with a simulation prior and converge toward the Bayes-optimal predictor, and (ii) they can provably recover unobservable scientific quantities (with interpretability tied to the mechanistic simulation) even when empirical methods fail.
Published in arXiv, 2025
Introduced Mantis, a foundation model for infectious disease forecasting that outperformed all 39 expert tuned models it was tested against across 6 diseases (including all models in the CDC’s COVID-19 Forecast Hub).
Published in arXiv, 2025
Introduced Simulation-Grounded Neural Networks (SGNNs), a new modeling framework that uses mechanistic simulations to train neural networks, combining the interpretability of scientific theory with the flexibility of deep learning. SGNNs deliver state-of-the-art performance across disciplines—from forecasting epidemics to inferring hidden variables—while offering a novel form of mechanistic interpretability.
Published in Mathematical Biosciences and Engineering, 2022
Developed a method using topological data analysis to track and quantify dynamic structural changes in cellular filament networks over time, enabling new insights into biological processes like wound healing and the effects of experimental perturbations.
Published in Journal of the American Heart Association, 2019
Analyzed heart failure hospitalizations in adults with congenital heart disease, finding they had longer stays, more complications, and higher in-hospital death rates compared to patients without congenital heart disease—highlighting the need for tailored care protocols.