Gibbs sampling with deterministic dependencies

Publikationstyp
Conference Paper
Date Issued
2011-12-26
Sprache
English
Author(s)
Gries, Oliver  
Institut
Softwaresysteme E-16  
TORE-URI
http://hdl.handle.net/11420/11541
First published in
Lecture notes in computer science  
Number in series
7080 LNAI
Start Page
418
End Page
427
Citation
5th international workshop, MIWAI 2011, Hyderabad, India, December 7 - 9, 2011 ; proceedings. - (Lecture Notes in Computer Science ; Vol. 7080 LNAI). - Seite 418-427 (2011)
Contribution to Conference
5th International Workshop MIWAI 2011  
Publisher DOI
10.1007/978-3-642-25725-4_37
Scopus ID
2-s2.0-84055189601
Publisher
Springer
There is a growing interest in the logical representation of both probabilistic and deterministic dependencies. While Gibbs sampling is a widely-used method for estimating probabilities, it is known to give poor results in the presence of determinism. In this paper, we consider acyclic Horn logic, a small, but significant fragment of first-order logic and show that Markov chains constructed with Gibbs sampling remain ergodic with deterministic dependencies specified in this fragment. Thus, there is a new subclass of Gibbs sampling procedures known to approximate the correct probabilities and expected to be useful for lots of applications.
Subjects
Deterministic Dependencies
Gibbs Sampling
Markov Networks
Probabilistic Inference
DDC Class
004: Informatik