A non-parametric hierarchical clustering model

We present a novel non-parametric clustering model using Gaussian mixture model (NHCM). NHCM uses a novel Dirichlet process (DP) prior allowing for more flexible modeling of the data, where the base distribution of DP is itself an infinite mixture of Gaussian conjugate prior. NHCM can be thought of as hierarchical clustering model, in which the low level base prior governs the distribution of the data points forming sub-clusters, and the higher level prior governs the distribution of the sub-clusters forming clusters. Using this hierarchical configuration, we can maintain low complexity of the model and allow for clustering skewed complex data. To perform inference, we propose a Gibbs sampling algorithm. Empirical investigations have been carried out to analyse the efficiency of the proposed clustering model.