|
There is so much stuff coming out in the area of non-parametric bayesian methods, and usually Dirichlet Processes (DP) are involved. While there are lots of good tutorials on non-parametric bayesian methods, is there any good introduction to DPs? Edit: I'll add these non-parametric bayesian tutorials, though this might be a worthwhile topic on its own? Dirichlet processes, Chinese restaurant processes and all that, by M. Jordan Dirichlet Process Tutorial and Practical Course, by Y. W. Teh Non-parametric bayesian methods, by Z. Ghahramani |
|
The Dirichlet processes isn't new and itself dates back to the late 60's. There's actually a lot of great stuff out of the statistics literature that Mike Jordan's group was reading when a lot of this work was being developed. So reading the original papers is a good idea. I also actually recommend reading this paper that describes the DP from the statistical point-of-view. It does a good job explaining the multiple representations of the DP (e..g, CRP, limit of finite mixtures, or Stick-breaking representation). |
|
There's Michael Jordan's "Dirichlet Processes, Chinese Restaurant processes and all that" talk. Besides that, I think Neal's paper on sampling for dirichlet process mixtures, Blei and Jordan's variational inference paper, Yee Whye Teh's encyclopedia article on DPs, and Teh et al's HDP paper. I never found a single one of these sufficient, but all of them together have answered all questions I had on DPs. +1 for Y.W. Teh's encyclopedia article.
(Jul 04 '10 at 08:07)
osdf
|
|
While I learnt a lot about DPs going through the tutorials for non-parametric bayesian methods, looking at code and implementing stuff on my own, I always wanted to have some 'nicely' written intro. By chance I recently found Eric Sudderth's PhD thesis, which has a nice introductory section (2.5) on DPs. |
|
Emily Fox also has a great introduction to Bayesian nonparametric methods in her thesis [1]. I've used that as a reference on multiple occasions. Her background section gives an almost textbook-like treatment of Graphical Models, HMMs, and Bayesian nonparametrics. It's pretty awesome. [1] - http://www.mit.edu/~ebfox/publications/ebfox_thesis.pdf |
|
i like this paper for first learning about dirichlet processes ( navarro et. al. - modeling individual differences using the dirichlet process), as it's written for a psychological audience and the author is a good writer. that being said, i am a psychologist, so i may be biased :) |
|
Y. W. Teh has a good intro, although he kinda delves a bit in unrelated stuff here: You might try to go through his lecture, is quite good, specially the part 2. |
|
I found Volker Tresp's tutorial from MLSS'06 also to be a good introduction into DPs: Dirichlet Processes and Nonparametric Bayesian Modelling (video) |
