Algorithms for Markov Random Trees: marginal probabilities,                   
learning, tuning.                                                             
                                                                               
 Alexander Shekhovtsov                                                         
                                                                               
 International Research and Training Center for Information                    
 Technologies and Systems, Ukrainian Akademy of Sciences,                      
 Kiev                                                                          
                                                                               
 Abstract                 

 Consider a Markov Chain Model able to branch at random. An                    
 instance of this branching process is a random tree of                        
 hidden states. Such a tree of hidden states can naturally                     
 describe a tree-structured object, for instance a                             
 mathematical formula. Recognition based on maximal a                          
 posteriori probability, calculation of the marginal                           
 probability and selection of model parameters for such                        
 structures will be addressed in the talk. Algorithms                          
 solving above mentioned problems are known for Markov                         
 chains. The aim of this work is to transfer algorithms to                     
 the tree model.