Adaptive exact-approximate sequential Monte Carlo

Exact-approximate sequential Monte Carlo (SMC) methods target the exact posterior of intractable likelihood models by using a non-negative unbiased estimator of the likelihood when the likelihood is computationally intractable. For state-space models, a particle filter estimator can be used to obtain an unbiased estimate of the likelihood. The efficiency of exact-approximate SMC greatly depends on the variance of the likelihood estimator, and therefore on the number of state particles used within the particle filter. We introduce a novel method to adaptively select the number of state particles within exact-approximate SMC. We also utilise the expected squared jumping distance to trigger the adaptation, and modify the exchange importance sampling method of Chopin et al. (2012) to replace the current set of state particles with the new set. The resulting algorithm is fully adaptive, and can significantly improve current methods. Code for our methods is available at https://github.com/imkebotha/adaptive-exactapproximate-smc. Keywords— Bayesian inference, State-space models, SMC, Pseudo-marginal, Particle MCMC

[1]  Paul Fearnhead,et al.  An Adaptive Sequential Monte Carlo Sampler , 2010, 1005.1193.

[2]  Pierre E. Jacob,et al.  Path storage in the particle filter , 2013, Statistics and Computing.

[3]  G. Peters,et al.  Ecological non-linear state space model selection via adaptive particle Markov chain Monte Carlo (AdPMCMC) , 2010, 1005.2238.

[4]  Robert Kohn,et al.  Particle Methods for Stochastic Differential Equation Mixed Effects Models , 2019, 1907.11017.

[5]  C. Andrieu,et al.  The pseudo-marginal approach for efficient Monte Carlo computations , 2009, 0903.5480.

[6]  N. Gordon,et al.  Novel approach to nonlinear/non-Gaussian Bayesian state estimation , 1993 .

[7]  Haikady N. Nagaraja,et al.  Inference in Hidden Markov Models , 2006, Technometrics.

[8]  N. Chopin.  A sequential particle filter method for static models , 2002 .

[9]  P. Moral,et al.  Sequential Monte Carlo samplers , 2002, cond-mat/0212648.

[10]  Matteo Fasiolo,et al.  A comparison of inferential methods for highly nonlinear state space models in ecology and epidemiology , 2014 .

[11]  A. Gelman,et al.  Adaptively Scaling the Metropolis Algorithm Using Expected Squared Jumped Distance , 2007 .

[12]  Andrew Golightly,et al.  Ensemble MCMC: Accelerating Pseudo-Marginal MCMC for State Space Models using the Ensemble Kalman Filter , 2019, Bayesian Analysis.

[13]  Jin-Chuan Duan,et al.  Density-Tempered Marginalized Sequential Monte Carlo Samplers , 2013 .

[14]  James Ridgway,et al.  Towards automatic calibration of the number of state particles within the SMC$^2$ algorithm , 2015, 1506.00570.

[15]  Robert Kohn,et al.  Robustly estimating the marginal likelihood for cognitive models via importance sampling , 2019, Behavior Research Methods.

[16]  J. Rosenthal,et al.  On the efficiency of pseudo-marginal random walk Metropolis algorithms , 2013, The Annals of Statistics.

[17]  Dirk P. Kroese,et al.  Unbiased and consistent nested sampling via sequential Monte Carlo , 2018, 1805.03924.

[18]  B. Øksendal.  Stochastic differential equations : an introduction with applications , 1987 .

[19]  Ralph S. Silva,et al.  On Some Properties of Markov Chain Monte Carlo Simulation Methods Based on the Particle Filter , 2012 .

[20]  A. Doucet,et al.  Efficient implementation of Markov chain Monte Carlo when using an unbiased likelihood estimator , 2012, 1210.1871.

[21]  A. Doucet,et al.  Particle Markov chain Monte Carlo methods , 2010 .

[22]  Anthony Lee,et al.  Accelerating sequential Monte Carlo with surrogate likelihoods , 2020, Statistics and Computing.