HybridAlpha

Abstract

Federated learning has emerged as a promising approach for collaborative and privacy-preserving learning. Participants in a federated learning process cooperatively train a model by exchanging model parameters instead of the actual training data, which they might want to keep private. However, parameter interaction and the resulting model still might disclose information about the training data used. To address these privacy concerns, several approaches have been proposed based on differential privacy and secure multiparty computation (SMC), among others. They often result in large communication overhead and slow training time. In this paper, we propose HybridAlpha, an approach for privacy-preserving federated learning employing an SMC protocol based on functional encryption. This protocol is simple, efficient and resilient to participants dropping out. We evaluate our approach regarding the training time and data volume exchanged using a federated learning process to train a CNN on the MNIST data set. Evaluation against existing crypto-based SMC solutions shows that HybridAlpha can reduce the training time by 68% and data transfer volume by 92% on average while providing the same model performance and privacy guarantees as the existing solutions. CCS CONCEPTS• Security and privacy → Privacy-preserving protocols; • Computing methodologies → Distributed artificial intelligence; Neural networks.

Towards federated learning: An overview of methods and applications

“…However, this method results in high communications costs and long convergence times, such as HE. Then, the HybridAlpha emerged as a method that combines functional encryption with SMC protocol to achieve a highly‐performance model without privacy sacrifice (R. Xu et al, 2019).…”

Challenges, Applications and Design Aspects of Federated Learning: A Survey

“…Aledhari et al [102] mainly focus on architecture options for FL based models -Horizontal FL [89], Vertical FL [89], MMVFL [103], FTL [71], FEDF [104], PerFit [105], FedHealth [106], FADL [107], Blockchain-FL [108], whereas a primary focus of [109] is on aggregation techniques -FedAvg [1], SMC-avg [19], FedProx [4], FedMA [110], Scaffold: Stochastic Controlled Averaging for FL [58], Tensor Factorization [111], FedBCD [31], Federated Distillation (FD) and Federated Augmentation (FAug) [18], Co-Op, LoAdaBoost [17], HybridFL [34], FedCS [112], PrivFL [113], VerifyNet [114].…”

PROV-FL: Privacy-preserving Round Optimal Verifiable Federated Learning

“…In view of these attacks, PPML protocols have been developed. Existing techniques for designing PPML protocols can be broadly classified into four categories: 1) secure multi-party computation techniques, e.g., [3,11,13,28,29,33], 2) homomorphic encryption, e.g., [24,26,35], 3) differential privacy and homomorphic encryption or secure aggregation, e.g., [9,40,42], and 4) leveraging trusted execution environments (e.g., Intel-SGX), e.g., [18,30]. In the private training using secure multi-party computation, the training data is shared using a secret-sharing protocol among a small set of servers (e.g., 2, 3 or 4-server) (e.g., [11,28,29]), and then the training is conducted and the model is secret-shared among participating servers.…”

“…First, we propose a secure, verifiable and robust model aggregation protocol with single round in the presence of dynamic user participations or dropouts. Compared to multi-round secure aggregation solutions (e.g., [9,26]) and solutions that are aided by a fully trusted third party (TTP) (e.g., [42]), PROV-FL achieves a single-round ML model aggregation, without a fully TTP, that is robust against dropouts. Second, we construct the PROV-FL training protocols by combining our new model aggregation protocol in the federated training along with the differential privacy mechanism to provide a good balance between efficiency, privacy and accuracy in training an ML model.…”

“…Our results show that training in PROV-FL does not compromise users' privacy while achieving above 90% accuracy (with (0.5, 1×10 −5 )-DP noise) and reduced communication and computational complexities. Compared with HybridAlpha [42] which uses Multi Input Functional Encryption (MIFE) [2] for secure aggregation, PROV-FL's execution time is about 103 seconds per user for 118, 110 gradients while HybridAlpha takes 326 seconds per user for the MIFE encryption (using the CiFEr Project's MIFE implementation). A detailed comparison between PROV-FL and HybridAlpha is provided in Section 5.4.…”