1. Data driven Wiener system identification
2. ex) Particle Markov chain Monte Carlo
This small code package implements two Particle Markov chain Monte Carlo (PMCMC) methods for Bayesian parameter inference. The two algorithms are particle Gibbs with ancestor sampling (PGAS, [pdf]) and particle marginal Metropolis-Hastings (PMMH, [pdf]). The intention is to illustrate the algorithm on a simple example. The MATLAB code can be accessed here, [code].
3. ex) Stochastic approximation EM using conditional particle filters
This code package implements the CPF-SAEM algorithm (conditional particle filter stochastic approximation EM) for maximum likelihood identification of a nonlinear system (the same system as used in the Bayesian setting in code package 2; see above). The algorihtm is described in [pdf]. As comparison, a particle smoothing EM (PSEM) algorithm is also implemented. The main advantage of CPF-SAEM over PSEM is that it makes more efficient use of the simulated values, reducing the computational cost. The MATLAB code can be accessed here, [code].
Page responsible: Fredrik Lindsten
Last updated: 2014-03-26