KEYI LIU

Probabilistic Graphical Models

Learning Deep generative Causal model

Now, data is the most crucial source that we have at hand, so people try to find some patterns among the data, however, there might be some hidden causes that generate these data. Nowadays, vast majority of the deep models are used for discovering correlation relationship among random variables, like DBM and DBN. Since the correlation relationship does not imply the causal relationship, when there is prior knowledge indicating that causal relationship exists, correlation relationship discovery may not capture this property well. This motivates us to extract this causal relationship by constructing a deep causal network in which the NoBN is the building block. When learning the latent-node generative model, learning also needs inference. The problem for inference is that doing exact posterior inferece is intractable when network is large. So, my current research focuses on this part.