Particle Gibbs with ancestor sampling for probabilistic programs

Conference item

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.

Authors: Meent, J; Yang, H; Mansinghka, V; Wood, F

• Publication status: Published
• Peer review status: Reviewed (other)
• Publisher: Journal of Machine Learning Research Publisher's website
• Host title: Proceedings of the 18th International Conference on Artificial Intelligence and Statistics (AISTATS)
• Journal: Proceedings of the 18th International Conference on Artificial Intelligence and Statistics (AISTATS Journal website
• Volume: 38
• Pages: 986-994
• Publication date: 2015-02-21
• Acceptance date: 2015-01-11
• EISSN: 1938-7228
• ISSN: 2640-3498