Three Layers Latent Tree Model

Latent Tree Model is a tree structure of Bayesian network model contains latent variables,are significant variables in the middle of the Bayesian network nodes, internal nodes are latentvariables.[1]This paper mainly through the structure of the model is improved in this area,three layers latent tree model was first proposed three-layers structure. The three layers latenttree model retains the advantages of the original tree model analysis variables implicit internalrelations, and have the model structure concise, strong interpretability, the algorithm speedcharacteristics, especially the model interpretability is this paper concern, but also put forwardthree main significance lies hidden tree model.Implicit in the triple tree construction algorithm model, we propose two types ofinformation from discrimination law and an effective means of mutual information packetmethod. Information from or through mutual information we can be divided into differentvariables significantly teams, each introducing a latent variable constitute an implicit classmodel for each hidden class model for us to learn the model parameters via the EM algorithm,by BIC scoring function to determine the number of latent variables state. Finally, a root node(a latent variable) will put together these groups constitute an implicit model of the three-tierstructure of the tree, then studied parameters via the EM algorithm to determine the root of thenumber of states by BIC scoring function. Through a case study, I did not see three layerslatent tree model does have a structure concise, fast, interpretability and strong features.