Automated hypothesis generation via Evolutionary Abduction

Roberto Pietrantuono

Keywords: Causal inference, abduction, evolutionary computation, learning

Abstract: Abduction is a powerful form of causal inference employed in many artificial intelligence tasks, such as medical diagnosis, criminology, root cause analysis, intent recognition. Given an effect, the abductive reasoning allows advancing a plausible set of explanatory hypotheses for its causes. This paper presents a new evolutionary strategy - called Evolutionary Abduction (EVA) - for automated abductive inference, aiming at effectively generating sets of hypotheses for explaining an occurred effect and/or predicting an effect that could occur in the future. EVA defines a set of abductive operators to repeatedly construct hypothetical cause-effect instances, and then automatically assesses their plausibility as well as their novelty with respect to already known instances - a mechanism mimicking the human reasoning employed whenever we need to select the best candidates from a set of hypotheses. Experiments with four datasets confirm that, given a background knowledge, EVA can construct new and realistic multiple-cause hypotheses for a given effect. EVA outperforms alternative strategies based on causal structure discovery, generating closer-to-real instances in most settings and datasets.