Possibilistic Causal Networks for Handling Interventions: A New Propagation Algorithm

This paper contains two important contributions for the development of possibilistic causal networks. The first one concerns the representation of interventions in possibilistic networks. We provide the counterpart of the ”DO” operator, recently introduced by Pearl, in possibility theory framework. We then show that interventions can equivalently be represented in different ways in possibilistic causal networks. The second main contribution is a new propagation algorithm for dealing with both observations and interventions. We show that our algorithm only needs a small extra cost for handling interventions and is more appropriate for handling sequences of observations and interventions.