Weighted Model Integration

Explore weighted model integration (WMI), a framework for advanced probabilistic inference in hybrid domains, through this 48-minute lecture by Zhe Zeng from UCLA. Delve into the complexities of distributions over mixed continuous-discrete random variables and their interaction with logical and arithmetic constraints. Gain insights into existing WMI solvers and understand the tractability analysis of WMI problem classes, focusing on complexity characterization using primal graphs with treewidth and diameter. Discover the largest known tractable WMI problem class to date. Learn about the practical application of WMI in Bayesian deep learning, particularly in developing accurate uncertainty estimation algorithms.