Abstract
A causal model describes the distribution of each variable Xas a function of the variables causing X, and of independent noise (representing other unknown factors influencing X- the " known unknowns ").
A causal model avoids the robustness shortcomings of models based on correlations between variables. It can also help decision-makers
decision-making, by predicting the effects of interventions (what happens to the state of a system if we intervene in its functioning ?) and by counterfactual reasoning (what would have happened to the system if we had carried out intervention Ainstead of B ?)
The presentation presents causal modeling : the royal road (randomized controlled trials) and approaches for inferring a model from data.
