Particle Gibbs with Ancestor Sampling for Probabilistic Programs

By Jan-Willem van de Meent, Hongseok Yang, Vikash K. Mansinghka, Frank D. Wood. 2015

33 citations recorded by the source corpus.

Particle Markov chain Monte Carlo techniques rank among current state-of-the-art methods for probabilistic program inference. A drawback of these techniques is that they rely on importance resampling, which results in degenerate particle trajectories and a low effective sample size for variables sampled early in a program. We here develop a formalism to adapt ancestor resampling, a technique that mitigates particle degeneracy, to the probabilistic programming setting. We present empirical results that demonstrate nontrivial performance gains.

Subfields: Artificial Intelligence, Probabilistic & Bayesian Methods

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 Probabilistic & Bayesian Methods

Citations in the corpus

Design and Implementation of Probabilistic Programming Language Anglican (2016, methodology): “Typically inference can be performed for any probabilistic program using one or more generic inference techniques provided by the system back end, such as Metropolis-Hastings [9, 29, 33], Hamiltonian Monte Carlo [23], expectation propagation [11], and extensions of Sequential Monte Carlo [15, 28,…”
Applications of Probabilistic Programming (Master's thesis, 2015) (2016, methodology): “In 2014 a general-purpose implementation of the particle Gibbs with ancestor sampling method (Lindsten et al., 2014) was introduced in (van de Meent et al., 2015) as an alternative engine for Anglican.”
Interacting Particle Markov Chain Monte Carlo (2016, methodology): “We present empirical results that show significant improvements in mixing rates relative to both non-interacting PMCMC samplers, and a single PMCMC sampler with an equivalent memory and computational budget.”
Output-Sensitive Adaptive Metropolis-Hastings for Probabilistic Programs (2015, methodology): “a Kalman smoothing problem) that was previously described in [12]”
Practical probabilistic programming with monads (2015, methodology): “In recent work, van de Meent et al. (2015) provide a formal operational semantics for a Church-like language to associate expressions with execution traces, so as to explain an inference method, Particle Gibbs with Ancestor Sampling, as opposed to the semantics of the Bayesian model.”
Probabilistic Programming in Anglican (2015, methodology): “Typically inference can be performed for any probabilistic program using one or more generic inference techniques provided by the system backend, such as Metropolis-Hastings [8,3,10], Hamiltonian Monte Carlo [7], expectation propagation [5], and extensions of Sequential Monte Carlo [9,4,6] methods.”