<div>The increasing pressure toward transparent, explainable and accountable reasoning in Artificial Intelligence has caused renewed attention to be given to causal models. While the royal road toward establishing causal relationships is based on randomized experiments, these might be subject to severe ethic or feasibility restrictions, or too costly in some domains. For this reason, learning causal models from pure observational data is becoming a hot topic in machine learning. The talk will present Causal Generative Neural Networks (CGNNs) aimed to learn functional causal models from observational data. CGNNs leverage conditional independencies and distributional asymmetries to discover bivariate and multivariate causal structures. CGNNs make no assumption regarding the lack of confounders, and learn a differentiable generative model of the data by using backpropagation. Extensive experiments show their good performances comparatively to the state of the art in observational causal discovery on both simulated and real data, with respect to cause-effect inference, v-structure identification, and multivariate causal discovery.</div>