论文信息 - EXCHANGEABLE RANDOM PRIMITIVES

EXCHANGEABLE RANDOM PRIMITIVES

Exchangeable random primitives (XRPs) are a crucial, but undocumented, class of constructs in the probabilistic programming language Church. We give a rigorous definition for the samplers of XRPs, and give a representation theorem characterizing their essential form and conditional independence structure. We place semantic questions surrounding exchangeable random primitives in the context of a line of research in programming languages involving purity and referential transparency.

Daniel M. Roy | Cameron E. Freer | L. Ackerman

[1] Amr Sabry,et al. What is a purely functional language? , 1998, Journal of Functional Programming.

[2] John Longley. When is a functional program not a functional program? , 1999, ICFP '99.

[3] Peter Sestoft,et al. Referential transparency, definiteness and unfoldability , 1990, Acta Informatica.

[4] Vikash K. Mansinghka,et al. A stochastic programming perspective on nonparametric Bayes , 2008 .

[5] Joshua B. Tenenbaum,et al. Church: a language for generative models , 2008, UAI.

[6] Vikash K. Mansinghka,et al. Natively Probabilistic Computation by Vikash , 2009 .

[7] Chung-chieh Shan,et al. Embedded Probabilistic Programming , 2009, DSL.

[8] Daniel M. Roy,et al. Posterior distributions are computable from predictive distributions , 2010, AISTATS.

[9] Sebastian Fischer,et al. Purely functional lazy nondeterministic programming , 2011, J. Funct. Program..

[10] Daniel M. Roy,et al. Computable de Finetti measures , 2009, Ann. Pure Appl. Log..

[11] Yura N. Perov,et al. Venture: a higher-order probabilistic programming platform with programmable inference , 2014, ArXiv.