pipette
ENEnglish

Resolving Structural Uncertainty in Stochastic Dynamical Systems via Ensemble-Based Bayesian Inference

Francesco Zambelli, Valeria D'Andrea, Manlio De Domenico

PreprintAfirmaciones fuertes, leer con cuidado

En palabras de los autores

Recovering the interaction structure of complex stochastic dynamical systems from sparse time-series observations is an inherently ill-posed inverse problem. The posterior over network configurations is high-dimensional, highly degenerate, and shaped by the non-equilibrium nature of the available data. To address this, we propose a fully Bayesian, ensemble-based inference framework. Our approach grounds a continuous-time Stochastic Differential Equation (SDE) likelihood formulation---acting mathematically as an Onsager-Machlup quasi-potential for non-equilibrium paths---within a non-parametric Minimum Description Length (MDL) structural prior. This framework rigorously penalizes model complexity and avoids kinetic overfitting by forcing continuous parameters into a finite vocabulary of discrete categories. The joint posterior is optimized using a block-coordinate scheme combining sub-quadratic Markov Chain Monte Carlo (MCMC) topological sampling with deterministic Nelder-Mead parameter optimization. Using highly non-linear macroscopic population dynamics as a benchmark, we demonstrate that inference requires non-stationary transient data to resolve causal information. Furthermore, we prove that propagating structural uncertainty via ensemble-based posterior analysis produces significantly more robust and reliable predictions for downstream perturbation outcomes than traditional single-point (MAP) estimates. The framework is highly generalizable, capturing fundamental topological constraints even under severe generative model mismatches, offering a robust tool for uncertainty-aware inference in complex stochastic systems.

Resultado principalEl resumen no menciona limitaciones.

Apareció: jueves, 15 de octubre. arXiv. Preprint, todavía sin revisión por pares.

Comentario de los autores: 14 pages article, 4 figures article, 5 pages SM, 1 figure SM