Probabilistic Graphical Models are widely used tools to study relationships and visualize interactions in large and complex data sets. They represent probability distributions as a graph with edges denoting conditional dependence relationships between random variables. These models have been studied and developed in computer science, probability, and statistics, with wide ranging applications in artificial intelligence, computer vision, physics, systems biology, and neuroscience, to name a few. This talk will review structural learning in probabilistic graphical models, which seeks to learn the unknown edges or conditional dependencies between random variables, and highlight new methods and theory from my research group for graph structural learning from big biomedical data. Specifically, I will discuss graph learning for non-Gaussian data and mixed data via data integration, for non-simultaneously recorded or non-aligned data via Graph Quilting, and for graph learning in the presence of latent variables. Additionally, I will present several applications and case studies using these approaches to make scientific discoveries in integrative genomics and neuroscience.
The MLxMed Seminar Series focuses on new algorithms, real-world deployment, and future trends in machine learning in medicine. It will feature prominent investigators who are developing and applying machine learning to biomedical discovery and in clinical decision support.
Hosted by the Department of Biomedical Informatics