Learning nonlinear state-space models using smooth particle-filter-based\n likelihood approximations

2017/11/29 by Andreas Svensson, Svensson, Andreas, Fredrik Lindsten +3
Computer Science · Engineering · #Bayesian Modeling and Causal Inference #Computation (stat.CO) #FOS: Computer and information sciences #FOS: Electrical engineering #Fault Detection and Control Systems #Systems and Control (eess.SY) #Target Tracking and Data Fusion in Sensor Networks #electronic engineering #information engineering

paper · pdf · doi:10.48550/arxiv.1711.10765

openalex publication_date 2017/11/29 · openalex created_date 2022/08/23 · openalex updated_date 2026/07/28

Abstract

When classical particle filtering algorithms are used for maximum likelihood\nparameter estimation in nonlinear state-space models, a key challenge is that\nestimates of the likelihood function and its derivatives are inherently noisy.\nThe key idea in this paper is to run a particle filter based on a current\nparameter estimate, but then use the output from this particle filter to\nre-evaluate the likelihood function approximation also for other parameter\nvalues. This results in a (local) deterministic approximation of the likelihood\nand any standard optimization routine can be applied to find the maximum of\nthis local approximation. By iterating this procedure we eventually arrive at a\nfinal parameter estimate.\n

Related

• Asymptotic Properties of Recursive Maximum Likelihood Estimation in Non-Linear State-Space Models
• An efficient particle-based method for maximum likelihood estimation in nonlinear state-space models
• A surrogate-based approach to nonlinear, non-Gaussian joint\n state-parameter data assimilation
• Learning state and proposal dynamics in state-space models using differentiable particle filters and neural networks
• Sigma-Point Filtering and Smoothing Based Parameter Estimation in Nonlinear Dynamic Systems