zkDFL: An efficient and privacy-preserving decentralized federated learning with zero-knowledge proof

TL;DR

The paper tackles trust and privacy concerns in decentralized federated learning by introducing zkDFL, a ZKP-based aggregator that proves the correctness of aggregation without exposing client data. It combines blockchain-managed smart contracts with Groth16 zkSNARKs to verify that the global update, computed as the average of selected local updates, is accurate, formalized as $w_{global} = \frac{1}{m} \sum_{i=1}^{m} w_i$, while protecting inputs via a SNARK-friendly hash. The system hashes client inputs with MiMC7 and proves that the same inputs were used and averaged, enabling verifiability on-chain without revealing sensitive data. Experimental results on wearable IoT data show strong verifiability and privacy protection, while gas costs are substantially reduced compared with traditional decentralized FL, highlighting the practical viability of zkDFL in large-scale, privacy-conscious deployments.

Abstract

Federated learning (FL) has been widely adopted in various fields of study and business. Traditional centralized FL systems suffer from serious issues. To address these concerns, decentralized federated learning (DFL) systems have been introduced in recent years. With the help of blockchains, they attempt to achieve more integrity and efficiency. However, privacy preservation remains an uncovered aspect of these systems. To tackle this, as well as to scale the blockchain-based computations, we propose a zero-knowledge proof (ZKP)-based aggregator (zkDFL). This allows clients to share their large-scale model parameters with a trusted centralized server without revealing their individual data to other clients. We utilize blockchain technology to manage the aggregation algorithm via smart contracts. The server performs a ZKP algorithm to prove to the clients that the aggregation is done according to the accepted algorithm. Additionally, the server can prove that all inputs from clients have been used. We evaluate our approach using a public dataset related to the wearable Internet of Things. As demonstrated by numerical evaluations, zkDFL introduces verifiability of the correctness of the aggregation process and enhances the privacy protection and scalability of DFL systems, while the gas cost has significantly declined.

Figures (10)

• Figure 1: MiMC7 Structure.
• Figure 2: Architecture of the proposed system.
• Figure 3: MLP Architecture.
• Figure 4: Weight combination.
• Figure 5: Accuracy for different Model Architectures and various number of clients.