Réseaux de Neurones Génératifs pour la Découverte de Mécanismes Causaux: Algorithmes et Applications
Diviyan Kalainathan · HAL (Le Centre pour la Communication Scientifique Directe) · 2019
Causal discovery is of utmost importance for agents who must plan, reason anddecide based on observations; where mistaking correlation with causation mightlead to unwanted consequences. The goldstandard to discover causal relations is to perform experiments.However, experiments are in many cases expensive, unethical, or impossible torealize. In these situations, there is a need for observational causaldiscovery, that is, the estimation of causal relations from observations alone. Causal discovery in the observational data setting traditionally involves making significant assumptions on the data and on the underlying causal model.This thesis aims to alleviate some of the assumptions made on the causal models by exploiting the modularity and expressivenessof neural networks for causal discovery, leveraging both conditionalindependences and simplicity of the causal mechanisms through two algorithms.Extensive experimentson both simulated and real-world data and a throughout theoretical anaylsisprove the good performance and the soundness of the proposedapproaches.