Large-Scale Differentiable Causal Discovery of Factor Graphs - Lecture and Primer

Explore cutting-edge research on large-scale causal discovery and factor directed acyclic graphs (f-DAGs) in this comprehensive seminar from the Broad Institute's Models, Inference and Algorithms series. Delve into Romain Lopez's presentation on Differentiable Causal Discovery of Factor Graphs (DCD-FG), a novel approach for analyzing high-dimensional interventional data. Learn about the challenges of causal discovery in large datasets and how f-DAGs can restrict the search space to non-linear low-rank causal interaction models. Discover the theoretical analysis of edge perturbations on f-DAG skeletons and their implications for high-dimensional causal discovery. Additionally, engage with Jiaqi Zhang's primer on causal representation learning of genetic perturbations, exploring identifiability and combinatorial extrapolation in the context of single-cell assays and CRISPR experiments. Gain insights into predicting the effects of unseen intervention combinations and the implementation of causal disentanglement frameworks using autoencoding variational Bayes. This seminar offers valuable knowledge for researchers and practitioners working in causal inference, computational biology, and machine learning.