| Epidemic monitoring generates large volumes of geographically distributed medical time-series data that are essential for calibrating epidemiological models. However, privacy regulations often prevent the centralization of such sensitive data across healthcare institutions. Federated learning provides a natural framework for collaborative modeling without sharing raw data. In this work, we study federated parameter inference for simulation-based epidemiological models using Approximate Bayesian Computation with Sequential Monte Carlo (ABC-SMC). We propose two Bayesian federated architectures—Federated Bayesian Averaging and Federated Consensus—that enable posterior inference a\-cross distributed data silos without gradient exchange, making them compatible with complex non-differentiable simulators such as EpiSim. Experiments on real COVID-19 mobility and epidemiological data from Spain show that the proposed approaches accurately recover epidemic parameters while preserving data locality and achieving significant computational speedups through distributed execution. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.