Particle Learning and Smoothing

By Carlos M. Carvalho, Michael Johannes, Hedibert Freitas Lopes, Nicholas G. Polson. 2010

358 citations recorded by the source corpus.

Particle learning (PL) provides state filtering, sequential parameter learning and smoothing in a general class of state space models. Our approach extends existing particle methods by incorporating the estimation of static parameters via a fully-adapted filter that utilizes conditional sufficient statistics for parameters and/or states as particles. State smoothing in the presence of parameter uncertainty is also solved as a by-product of PL. In a number of examples, we show that PL outperforms existing particle filtering alternatives and proves to be a competitor to MCMC.

Subfields: Artificial Intelligence, Control and Systems Engineering

View Paper: https://www.semanticscholar.org/paper/dd82d0bc7d7e9730d6b8726e89247289aa05b165

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Control and Systems Engineering

Citations in the corpus

KalmanNet: Neural Network Aided Kalman Filtering for Partially Known Dynamics (2021, methodology): “robust variations were proposed for various state estimation algorithms, including Kalman variants [15]–[17], [53] and particle filters [54], [55].”
Variable selection and sensitivity analysis using dynamic trees, with an application to computer code performance tuning (2011, methodology): “This is achieved through use of a ﬁltering algorithm that follows a general particle learning recipe set out by Carvalho et al. (2010).”
Dynamic Trees for Learning and Design (2009, positive): “t 1 jzt 1) !p([S;˚]t 1 jzt) introduces information from ztthat is not conditional on ˚t, thus reducing direct dependence of particles 14 on the high dimensional ˚t. Further discussion is available in Carvalho et al. (2009b), which focuses on general mixtures and thus describes a partition model with”
SMC2: an efficient algorithm for sequential analysis of state space models (2011, methodology): “on-line algorithm may be obtained by extending SOPF so as to include MCMC updates of the -component, see Gilks and Berzuini (2001), Fearnhead (2002), Storvik (2002), and also the more recent work of Carvalho et al. (2010), but numerical evidence seems to indicate these algorithms degenerate as well,”
Bayesian inference for palaeoclimate with time uncertainty and stochastic volatility (2015, methodology): “For similar models there are numerous algorithms based on sequential Monte Carlo sampling (see, for example, Carvalho et al. (2010) for a review).”
Sequential Learning, Predictability, and Optimal Portfolio Returns: Sequential Learning, Predictability, and Optimal Portfolio Returns (2014, positive): “Particle filters discretize the support of the posterior, and, as shown by Johannes and Polson (2006) and Carvalho et al. (2010, 2011), work well for parameter and state variable inference in many models with latent states such as log-SV models.”
Online Context-Aware Recommendation with Time Varying Multi-Armed Bandit (2016, methodology): “Especially in the UPDATE procedure, we use the resample-propagate strategy in particle learning [6] rather than the propagate-resample strategy in particle filtering [10].”
Optimization Under Unknown Constraints (2010, methodology): “They show how GP and CGP models can be implemented in an online setting, by quickly updating a discrete approximation to the posterior via particle learning (Carvalho et al., 2008).”
Autoregressive Mixture Models for Dynamic Spatial Poisson Processes: Application to Tracking Intensity of Violent Crime (2010, methodology): “The particle learning updates of Carvalho et al. (2010) are p(St,Dt, ηt|yt) = ∫ p(St,Dt, ηt|yt,ψ,St−1,Dt−1)dP(ψ,St−1,Dt−1|yt) ∝ ∫ p(St,Dt, ηt|yt,ψ,St−1,Dt−1) × p(yt|St−1,ψ)dP(ψ,St−1,Dt−1|yt−1), (6) where further factorization yields p(ψ,St−1,Dt−1|yt−1) = p(ψ |Dt−1)p(St−1,Dt−1|yt−1) and p(St,Dt,…”
Sequential quasi Monte Carlo (2015, methodology): “Similarly, the trick at the end of §4.2.1, using sqMC on a single instead of (t+ 1) vectors, is unclear, but I wonder at a connection with the particle learning literature (Lopes et al., 2010; Carvalho et al., 2010).”