# Causal Discovery

The problem of learning the causal graph representing the system of interest using data. Differs from Causal Representation Learning, where the problem setting focuses on latent variable models.

Under some assumptions, we can rule out the existence of latent confounders, e.g. if we have \(X_1\to X_3\to X_4\) and the conditional independence relation \(X_1,X_4\) being independent conditional on \(X_3\), then we know there cannot be a latent confounder over \(X_3,X_4\) since if it exists, conditioning on \(X_3\) would contradict the conditional independence relation.