Speculative Computation with Multi-Agent Belief Revision

摘要

In multi-agent system, we often face incompleteness of information due to communication failure or other agent's suspension of decisions. To solve the incompleteness, we previously proposed speculative computation using abduction in the context of matter-slave multi-agent systems and gave a procedure in abductive logic programming. In the work, a master agent prepares a default value for a question in advance and it performs speculative computation using the default without waiting for a reply for the question. This computation is effective unless the contradictory reply with the default is returned. However, we find that this mechanism is not sufficient for speculative computation in more general multi-agent systems such that replies can be revised according to other agents' speculative computation. In this paper, we formalize speculative computation with multi-agent belief revision and propose a correct procedure for such computation.