Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of GraphSearch.
This lecture by the instructor covers the topic of causal discovery based on latent variable models. It delves into the fundamental problem of causal analysis, distinguishing between assumption-free and assumption-based methods. The lecture explores the challenges of identifying causal relationships in non-Gaussian data and the importance of considering temporal versus instantaneous effects. It introduces the concept of Linear non-Gaussian Acyclic Model (LiNGAM) and explains how non-Gaussianity can break the symmetry between variables, aiding in causal inference. The lecture also discusses the estimation of LiNGAM using Independent Component Analysis (ICA) and the difficulties and solutions in nonlinear Structural Equation Models (SEM) through methods like NonSENS. It concludes by emphasizing the significance of finding the direction of effect in causal discovery and the future prospects in this field.