Martina Rama: Physics-Informed Neural Networks for Probabilistic Epidemic Forecasting

Описание: Date: 12 June 2025
Speaker: Martina Rama
Title: Physics-Informed Neural Networks for Probabilistic Epidemic Forecasting
Abstract: Accurate epidemic forecasting is critical for informing public health decisions and timely interventions. While Physics-Informed Neural Networks (PINNs) have shown promise in various scientific domains, their application to real-time epidemic forecasting remains limited. The reasons are mainly due to the intrinsic difficulty of the task and the tendency to fully leveraging their learning and inference potential, which, however, often results in non-optimal forecasting frameworks. The first part of the talk will offer a general overview of Physics-Informed Neural Network hybrid methodology and compartmental epidemic models, setting the stage for the second part, which will focus on our proposed PINN-based framework for real-time epidemic forecasting. SIR-INN framework integrates the mechanistic structure of the classical Susceptible- Infectious-Recovered (SIR) model into a neural network architecture. Trained once on synthetic epidemic scenarios, the model is able to generalize across epidemic conditions without retraining. From limited and noisy observations, SIR-INN infers key transmission parameters via Markov chain Monte Carlo (MCMC) generating probabilistic short- and long-term forecasts with built-in uncertainty quantification. The talk will conclude with the presentation of validation results on seasonal influenza data provided by the Italian National Institute of Health, covering the 2023–2024 and 2024–2025 seasons. The proposed framework performs competitively with current state-of-the-art models, especially in terms of Mean Absolute Error (MAE) and Weighted Interval Score (WIS). It delivers accurate predictions across different phases of the outbreak, with improved performance observed in the latest season. Credible uncertainty intervals are consistently maintained, despite occasional shortcomings in coverage. Overall, SIR-INN emerges as a computationally efficient, interpretable, and generalizable solution for real-time epidemic forecasting.