Distributed combination of belief functions

We consider the problem of combining belief functions in a situation where pieces of evidence are held by agents at the node of a communication network, and each agent can only exchange information with its neighbors. Using the concept of weight of evidence, we propose distributed implementations of Dempster's rule and the cautious rule based, respectively, on average and maximum consensus algorithms. We also describe distributed procedures whereby the agents can agree on a frame of discernment and a list of supported hypotheses, thus reducing the amount of data to be exchanged in the network. Finally, we show the feasibility of a robust combination procedure based on a distributed implementation of the random sample consensus (RANSAC) algorithm.

Automated summary

This study addresses the combination of belief functions in a communication network where agents hold evidence and can only exchange information with neighbors. Distributed implementations of Dempster's rule and the cautious rule are proposed based on average and maximum consensus algorithms. Procedures for agents to agree on a frame of discernment and supported hypotheses are described to reduce data exchange, along with a demonstration of the feasibility of a robust combination procedure using a distributed implementation of the RANSAC algorithm.