Active Simulation-Based Inference of Coupled Oscillator Dynamics

Inferring microscopic parameters that give rise to emergent collective behavior is a central problem in nonequilibrium physics. We study nonlinear stochastic oscillator networks that exhibit synchronization and transport under noise and external forcing. To recover the statistics of microscopic couplings from coarse observables, we introduce an adaptive inference framework that refines its sampling of parameter space based on dynamical sensitivity. Unlike conventional simulation-based inference, the method leverages response information to direct computational effort toward the most informative regions. This approach not only improves efficiency but also clarifies how macroscopic observables encode the structure of microscopic interactions, providing a data-driven route to uncover governing principles in driven stochastic systems.