Lecture
This lecture by the instructor covers the topics of model selection and local geometry in causal models. It discusses the challenges of inferring causality from observational data, the concept of causal structure learning, and the use of graphical models in model selection. The lecture explores undirected Gaussian graphical models, directed graphical models, and directed acyclic graphs. It delves into the difficulties of selecting models in discrete directed acyclic graphs and the implications of k-equivalence. The presentation also touches on the statistical consequences of k-near-equivalence and the computational consequences of model overlap. The instructor concludes by discussing the lack of convexity in certain models and proposes methods to leverage the closeness of true models for improved selection.