Byzantine-Resistant Secure Aggregation for Federated Learning Based on Coded Computing and Vector Commitment

Abstract

In this paper, we propose an efficient secure aggregation scheme for federated learning that is protected against Byzantine attacks and privacy leakages. Processing individual updates to manage adversarial behavior, while preserving privacy of data against colluding nodes, requires some sort of secure secret sharing. However, communication load for secret sharing of long vectors of updates can be very high. To resolve this issue, in the proposed scheme, local updates are partitioned into smaller sub-vectors and shared using ramp secret sharing. However, this sharing method does not admit bi-linear computations, such as pairwise distance calculations, needed by outlier-detection algorithms. To overcome this issue, each user runs another round of ramp sharing, with different embedding of data in the sharing polynomial. This technique, motivated by ideas from coded computing, enables secure computation of pairwise distance. In addition, to maintain the integrity and privacy of the local update, the proposed scheme also uses a vector commitment method, in which the commitment size remains constant (i.e. does not increase with the length of the local update), while simultaneously allowing verification of the secret sharing process.

Byzantine-Resilient Secure Federated Learning on Low-Bandwidth Networks

“…The future directions of BREA-SV are summarized below: One promising direction is reducing the sizes of commitments by preparing commitments for aggregation of individual local models' vectors similar to the study of Tayyebeh et al [32]. A research issue is how to prevent adversaries from generating contaminated aggregated vectors of which commitment does not pass the share verification.…”

“…The most related work with BREA-SV is [32], which extends BREA to improve the communication efficiency of the share verification of BREA. Reference [32] similarly performs the share verification, but reduces the size of each commitment.…”

“…The most related work with BREA-SV is [32], which extends BREA to improve the communication efficiency of the share verification of BREA. Reference [32] similarly performs the share verification, but reduces the size of each commitment. The idea of reduction is to verify the elements in each share collectively, without verifying each element one by one like BREA.…”

“…. , g β d ) (β ∈ F p ) introduced by [32] prevent this attack, but require a trusted entity to generate them. This attack weakens the Byzantine resistance of [32] to the same level as RoFL without the trusted entity.…”

“…, g β d ) (β ∈ F p ) introduced by [32] prevent this attack, but require a trusted entity to generate them. This attack weakens the Byzantine resistance of [32] to the same level as RoFL without the trusted entity. BREA-SV, on the other hand, has the same Byzantine resilience as BREA and achieves communication efficiency.…”