Distributed Discovery of Causal Networks in Pervasive Environments

Stefano Mariani, Franco Zambonelli

Abstract: In pervasive computing environments, learning the causal network of relationships between environmental variables is crucial to support situation recognition and planning. However, this may be impossible when computing nodes have only partial observability and control, as in the case of multiple fog nodes each sensing only a local portion of variables and controlling a local portion of actuators. In fact, the causal network learnt by individual nodes may not suffice to give them full awareness and control over the environment. In this paper, we propose a protocol for distributed causal discovery, where fog nodes in an environment cooperate with each other to expand their individual local knowledge of the causal network and acquire the minimal knowledge of the causal network needed to achieve awareness and control. We evaluate our approach in a smart home scenario, showing its superior performance with respect to global causal learning with full observability.