Joseph Lemaitre: Influpaint, a denoising diffusion model for forecasting US seasonal influenza

Speaker: Joseph Lemaitre

 

Institution: University of North Carolina at Chapel Hill

 

Title: Influpaint, a denoising diffusion model for forecasting US seasonal influenza

 

Abstract: Forecasting infectious disease dynamics is critical for timely public health interventions, yet it remains challenged by the non-stationarity and complex dependencies of epidemic systems. Current mechanistic and statistical approaches often struggle to capture multimodal uncertainty or emergent trends in limited data environments. Here, we introduce Influpaint, a generative framework that adapts Denoising Diffusion Probabilistic Models (DDPMs)—state-of-the-art in image synthesis—for epidemic forecasting. By encoding influenza seasons as spatiotemporal images where pixel intensity reflects incidence, Influpaint learns a rich distribution of disease dynamics from a hybrid dataset of surveillance and simulated trajectories. Forecasting is treated as a conditional generation (inpainting) task, allowing the model to project coherent futures based on partial observations without explicit mechanistic equations. We show that Influpaint generates realistic, diverse epidemic trajectories and achieves predictive accuracy competitive with the leading multi-model ensemble. In prospective real-time evaluation during the 2023–2025 U.S. CDC FluSight challenges, the model demonstrated robust performance, ranking first in accuracy among individual contributing systems for the 2024–2025 season. These results demonstrate that deep generative models can capture the intricate structure of infectious disease propagation, offering a flexible, data-driven paradigm for real-time public health forecasting.

 

Bio: Joseph Lemaitre is an Assistant Professor at the University of North Carolina at Chapel Hill and a researcher in infectious disease dynamics within the Atlantic Coast Center for Infectious Disease Dynamics and Analytics (ACCIDDA). He develops models that combine mechanistic and statistical methods, including generative AI, to support policy decisions. His work centers on COVID-19, seasonal influenza, and cholera. He earned a BSc in robotics, an MSc in computational mathematics, and a PhD in environmental engineering from EPFL, with doctoral work on cholera control strategies.