Abstract
dc:descriptionThe Gibbs sampler is one of the most popular algorithms for inference in statistical models. In this thesis, we introduce a herding variant of this algorithm that is entirely deterministic. We demonstrate, with simple examples, that herded Gibbs exhibits better convergence behavior for approximating the marginal distributions than Gibbs sampling. In particular, image denoising exemplifies the effectiveness of herded Gibbs as an inference technique for Markov Random Fields (MRFs). Also, we adopt herded Gibbs as the inference engine for Conditional Random Fields (CRFs) in Named Entity Recognition (NER) and show that it is competitive with the state of the art. The conclusion is that herded Gibbs, for graphical models with nodes of low degree, is very close to Gibbs sampling in terms of the complexity of the code and computation, but that it converges much faster.
Degree
thesis:*- Name thesis:degree_name
- Master of Science - MSc
- Level thesis:degree_level
- master's
- Discipline thesis:degree_discipline
- Computer Science
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2012
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Fang, Jing
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivs 3.0 Unported
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/43188