Workshop: Information-Theoretic Methods for Rigorous, Responsible, and Reliable Machine Learning (ITR3)
Coded Privacy-Preserving Computation at Edge Networks
Elahe Vedadi · Yasaman Keshtkarjahromi · Hulya Seferoglu
Multi-party computation (MPC) is promising for privacy preserving machine learning algorithms at edge networks, like federated learning. Despite their potential, existing MPC algorithms fail short of adapting to the limited resources of edge devices. A promising solution, and the focus of this work, is coded computation, which advocates the use of error-correcting codes to improve the performance of distributed computing through “smart” data redundancy. In this paper, we design novel coded privacy-preserving computation mechanisms; MatDot coded MPC (MatDot-CMPC) and PolyDot coded MPC (PolyDot-CMPC) by employing recently proposed coded computation algorithms; MatDot and PolyDot. We take advantage of the “garbage terms” that naturally arise when polynomials are constructed in the design of MatDot-CMPC and PolyDot-CMPC to reduce the number of workers needed for privacy-preserving computation.