Social-Aware Decentralization for Secure and Scalable Multi-party Computations

This work studies the problem of MPC decentralization - that is, identifying a set of computing nodes to securely and efficiently execute the multi-party computation protocol (MPC) over a sensitive dataset. To balance between underdecentralization with high risk and over-decentralization with high cost, our unique approach is to add social-awareness, that is, the MPC protocol, running over a social network, is properly decentralized among the computing nodes selected carefully based on their social relationship. The key technical challenge is in estimating the risk of collusion between nodes on whom the computation is run. We propose solutions to estimate the risk of collusion based on (incomplete) social relationship, as well as algorithms for finding the MPC nodes such that the risk of collusion is minimized. We evaluate our methods on several real-world network datasets, and show that they are effective in minimizing the risk levels. This work has potential in enabling efficient privacy-preserving data sharing and computation in emerging big-data federation platforms, in healthcare, financial marketplaces, and other application domains.