Decentralized Learning of Bayesian Networks from Private Data with Applications to Global Pandemic

Reasoning under conditions of uncertainty is important in many areas where posterior knowledge depends on prior likelihood estimation from data. Distributed computing can provide a leverage for enhancing Bayesian network (BN) structure learning while keeping data private to users. We propose a decentralized learning framework based on distributed computing of BNs in local sites. Our method yields a higher fitness score (FS) for BNs in local sites, preserves local data privacy and significantly reduces the computation cost by 69.09% in comparison with a centralized approach which yields a low FS score, and does not keep user data private.