Prof. Tsamardinos talk at LINCS (Laboratory of Information, Networking and Communication Sciences), Paris, France 24/02/2016.
Abstract: Computational Causal Discovery aims to induce causal models,causal networks, and causal relations from observational data withoutperforming or by performing only few interventions (perbutations,manipulations) of a system. While predictive analytics create models thatpredict customer behavior for example, causal analytics create models thatdictate how to affect customer behavior. A recent approach to causaldiscovery, which we call logic-based integrative causal discovery, will bepresented. This approach is more robust to statistical errors, makes morerealistic and less restrictive assumptions (e.g., admits latent confoundingfactors and selection bias in the data) and accepts and reasons withmultiple heterogeneous datasets that are obtained under different samplingcriteria, different experimental conditions (perbubations, interventions),and measuring different quantities (variables). The approach significantlyextends causal discovery based on Bayesian Networks, the simplest causalmodel available, and is much more suitable for real business or scientificdata analysis.