papers: Estimation of Dirichlet Distribution Parameters

Recently, I am interested in apply Pachinko Allocation topic models to the object recognition problems. Mixtures of Hierarchical Topics with Pachinko Allocation, ICML 2007 mentioned several methods in training the hPAM model, and here are the related papers: 

• Gibbs EM: 
• An introduction to MCMC for machine learning, Machine Learning 2003
• implementations of the Monte Carlo EM Algorithm, 2001
  
• fixed point iteration method for estimating the Dirichlet parameters: 
• Estimating a Dirichlet distribution, Minka 2000, and code
• Maximum Likelihood Estimation of Dirichlet Distribution Parameters by Jonathan Huang

papers: Pachinko Allocation

I found a few papers about Pachinko Allocation (PA) model. Basically, this is an enhanced LDA model: in LDA model, topics are assumed to be independent, while PA introduced a tree-structure to capture the correlations and hierarchies of the topics. It sounds a reasonable assumption. This model has been extended to non-parametric version and to object recognition.


Pachinko Allocation (PA) model:

• Pachinko Allocation: DAG-Structured Mixture Models of Topic Correlations ICML 2006
• Mixtures of Hierarchical Topics with Pachinko Allocation, ICML 2007
• Pachinko Allocation: Scalable Mixture Models of Topic Correlations, JMLR 2008

Extension of PA to non-parametric model:

• Nonparametric Bayes Pachinko Allocation, UAI 2007

Applications in object recognition:

• Object Recognition Based on Dependent Pachinko Allocation Model

Code of PAM at Wei Li's webpage