Distributed Computation: Privacy, Straggler Mitigation, and Security Against Colluding Workers

Abstract

In a distributed computation system, there is a master node who wants to compute a function of its own data by distributing the computation amongst several worker nodes. In distributed computation system considered here, the master wants to multiply two matrices, one owned by the master and other matrix is in the library of matrices shared by the worker nodes. The master divides the computation task among these nodes. After recovering the desired computation, any set of up to T cottuding workers should not know which matrix in the library was desired by the master, referred to as demand privacy and also do not have any information about master’s matrix, referred to as data privacy. In this paper, a distributed computing scheme is proposed that simultaneously ensures (1) data and demand privacy of the master against colluding workers (2) straggler mitigation and (3) security against malicious workers. Similar matrix multiplication scenario was considered against non-colluding worker nodes by M. Kim and J. Lee in Private Secure Coded Computation.