Model Streaming for Distributed Multi-Context Systems

Multi-Context Systems (MCS) are instances of a nonmonotonic formalism for interlinking heterogeneous knowledge bases in a way such that the information flow is in equilibrium. Recently, algorithms for evaluating distributed MCS have been proposed which compute global system models, called equilibria, by local computation and model exchange. Unfortunately, they suffer from a bottleneck that stems from the way models are exchanged, which limits the applicability to situations with small information interfaces. To push MCS to more realistic and practical scenarios, we present a novel algorithm that computes at most k ≥ 1 models of an MCS using asynchronous communication. Models are wrapped into packages, and contexts in an MCS continuously stream packages to generate at most k models at the root of the system. We have implemented this algorithm in a new solver for distributed MCS, and show promising experimental results.