|Title:||A Hamiltonian Monte Carlo method for boosting Bayesian parameter inference of stochastic differential equation models|
|Authors :||Ulzega, Simone|
|Conference details:||NDES 2017 - Conference on Nonlinear Dynamics Systems, Zernez, 5. - 7. Juni 2017|
|License (according to publishing contract) :||Not specified|
|Type of review:||Not specified|
|Subject (DDC) :||003: Systems|
|Abstract:||Parameter inference is a fundamental problem in data-driven modeling. Indeed, for making reliable probabilistic predictions, model parameters need to be calibrated to measured data and their uncertainty needs to be quantified. Bayesian statistics is a consistent framework where knowledge about parameters is expressed through probability distributions. These so-called posterior distributions can become computationally very expensive to evaluate, especially with non-trivial stochastic models. We present a novel Hamiltonian Monte Carlo method for boosting Bayesian parameter inference of nonlinear stochastic differential equation models. The algorithm relies on the reinterpretation of the posterior distribution as the partition function of a statistical mechanics system akin to a polymer. We thus reduce the Bayesian inference problem to simulating the dynamics of a fictitious linear molecule whose dynamics are confined by the data and the model. Our approach is very efficient, applicable to a wide range of inference problems and highly parallelizable.|
|Departement:||Life Sciences and Facility Management|
|Organisational Unit:||Institute of Applied Simulation (IAS)|
|Publication type:||Conference Other|
|Appears in Collections:||Publikationen Life Sciences und Facility Management|
Files in This Item:
There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.