| dc.description.abstract | Bayesian networks are widely adopted to model complex systems by characterizing their information into conditional independencies of 2 or more system variables. For example, Bayesian networks have been commonly used for identifying gene regulatory networks and modeling decision networks in machine learning. While being popular, the structure of a Bayesian network is usually unknown and has to be inferred from available data in most of the cases. To date, learning the structure of Bayesian networks is still a very challenging and nuanced task partly due to the non-identifiability issue of Bayesian networks, especially when the data are cross-sectional and observational. In this thesis, we are going to use simulation-based approaches to investigate precisely under what conditions a Bayesian network can be identifiable, and therefore recoverable, for cross-sectional observational data. We will also explore required assumptions and overall implications of our work. | en |
