collapsed Gibbs sampling
Summary
MCMC samplers can often be improved by marginalizing out a subset of the variables in closed form and performing MCMC over the remaining variables. This is more statistically efficient since each particle can cover a larger part of the distribution, and it can also improve mixing by allowing larger jumps.
Context
This concept has the prerequisites:
- Gibbs sampling
- MCMC convergence (Collapsed sampling is a way of improving mixing.)
- multivariate distributions (Collapsed Gibbs sampling involves marginalizing out some of the variables.)
Goals
- Be able to derive the update rules for collapsed Gibbs sampling
- Be aware of the motivations in terms of:
- greater statistical efficiency (from the Rao-Blackwell theorem)
- faster mixing
Core resources (we're sorry, we haven't finished tracking down resources for this concept yet)
Supplemental resources (the following are optional, but you may find them useful)
-Free-
→ Machine learning summer school: Markov chain Monte Carlo (2009)
-Paid-
→ Machine Learning: a Probabilistic Perspective
A very comprehensive graudate-level machine learning textbook.
Location:
Section 24.2.4, pages 841-844
→ Probabilistic Graphical Models: Principles and Techniques
A very comprehensive textbook for a graduate-level course on probabilistic AI.
Location:
Section 12.4, pages 526-532
Additional dependencies:
- importance sampling
- Bayesian networks
See also
-No Additional Notes-