[Paper Review] HybridAlpha: An Efficient Approach for Privacy-Preserving Federated Learning

Runhua Xu, Nathalie Baracaldo+3 more|arXiv (Cornell University)|Dec 12, 2019

Privacy-Preserving Technologies in Data41 references40 citations

TL;DR

HybridAlpha uses a multi-input functional encryption-based SM C in federated learning to reduce training time and communication while preserving model performance and privacy, even with dynamic participant participation.

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.

Motivation & Objective

• Motivate privacy-preserving federated learning and limit data leakage during model updates.
• Reduce communication overhead and training time compared to existing crypto-based SM C approaches.
• Support dynamic participant participation (dropouts and new joins) without re-keying.
• Provide strong privacy guarantees against honest-but-curious aggregators and colluding participants.
• Demonstrate applicability to CNNs on MNIST and compare against baseline SMC methods.

Proposed method

• Employ a multi-input functional encryption (MIFE) scheme to perform secure aggregation in FL.
• Use a trusted third party authority (TPA) to initialize keys and distribute per-participant public keys.
• Integrate a differential privacy mechanism with the MIFE-based aggregation.
• Introduce an Inference Prevention Module to limit key/attack vectors that could reveal individual updates.
• Allow dynamic participant participation without full re-keying by adjusting weighted aggregation vectors.
• Provide analysis and benchmarking against existing SMC approaches to quantify efficiency gains.

Experimental results

Research questions

• RQ1 Can HybridAlpha achieve privacy-preserving aggregation with fewer crypto-rounds than existing SMC-based FL approaches?
• RQ2 How does the functional encryption-based approach affect training time, communication cost, and model accuracy compared to TP-SMC and P-SMC baselines?
• RQ3 Does HybridAlpha robustly handle dropout and addition of participants during training?
• RQ4 What are the security/privacy guarantees under the defined threat model (honest-but-curious aggregator, colluding participants, trusted TPA)?

Key findings

• Training time is reduced by 68% on average compared to existing crypto-based SMC solutions.
• Data transfer volume is reduced by 92% on average.
• HybridAlpha achieves the same model performance and privacy guarantees as the existing approaches.
• The framework supports dynamic participant participation with dropout and re-joining without re-keying.
• An Inference Prevention Module mitigates inference attacks by curious aggregators and colluding participants.
• Evaluations on MNIST with a CNN demonstrate efficiency gains and robust privacy protections.