|
Abstract:
|
Sparse inverse covariance matrix modeling is a popular tool for learning conditional dependency structure between random variables. In this talk, we propose a novel Bayesian approach to learn multiple Gaussian graphical models. Unlike existing works, we assume data come from heterogeneous and unknown classes. Specifically, we address the structure learning problem by identifying the cluster first and then estimate multiple undirected networks in situations where some of the networks may be unrelated, while others share common features. Our main interests are graph topology and density estimation simultaneously. We devise an efficient Markov chain Monte Carlo algorithm to fit the model. Experimental results show that our method performs well on simulated networks. We illustrate the advantage of our model using breast cancer gene expression data.
|
