Academic
Publications
Particle Filtered MCMC-MLE with Connections to Contrastive Divergence

Particle Filtered MCMC-MLE with Connections to Contrastive Divergence,Arthur U. Asuncion,Qiang Liu,Alexander T. Ihler,Padhraic Smyth

Particle Filtered MCMC-MLE with Connections to Contrastive Divergence   (Citations: 1)
BibTex | RIS | RefWorks Download
Learning undirected graphical models such as Markov random fields is an important machine learning task with applications in many domains. Since it is usually intractable to learn these mod- els exactly, various approximate learning tech- niques have been developed, such as contrastive divergence (CD) and Markov chain Monte Carlo maximum likelihood estimation (MCMC-MLE). In this paper, we introduce particle filtered MCMC-MLE, which is a sampling-importance- resampling version of MCMC-MLE with addi- tional MCMC rejuvenation steps. We also de- scribe a unified view of (1) MCMC-MLE, (2) our particle filtering approach, and (3) a stochas- tic approximation procedure known as persistent contrastive divergence. We show how these ap- proaches are related to each other and discuss the relative merits of each approach. Empiri- cal results on various undirected models demon- strate that the particle filtering technique we pro- pose in this paper can significantly outperform MCMC-MLE. Furthermore, in certain cases, the proposed technique is faster than persistent CD.
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.
Sort by: