ZK-HybridFL: Zero-Knowledge Proof-Enhanced Hybrid Ledger for Federated Learning

TL;DR

ZK-HybridFL tackles scalability, security, and privacy in decentralized federated learning by integrating a DAG-based ledger with dedicated sidechains and zero-knowledge proofs. The framework uses ZKPs to privately validate local inferences, a loss-aware DAG for robust model aggregation, and an oracle-assisted sidechain to manage event-driven smart contracts, ensuring only verified updates influence the global model. Key contributions include the DAG with ZKP-gated admission, a set of five EDSCs for validation, submission, challenges, aggregation, and rewards, and extensive simulations showing faster convergence, higher accuracy, and resilience to adversarial and idle nodes compared with Blade-FL and ChainFL. The results demonstrate sub-second on-chain verification, efficient gas usage, and robustness against orphanage attacks, making decentralized FL more scalable and secure for diverse environments.

Abstract

Federated learning (FL) enables collaborative model training while preserving data privacy, yet both centralized and decentralized approaches face challenges in scalability, security, and update validation. We propose ZK-HybridFL, a secure decentralized FL framework that integrates a directed acyclic graph (DAG) ledger with dedicated sidechains and zero-knowledge proofs (ZKPs) for privacy-preserving model validation. The framework uses event-driven smart contracts and an oracle-assisted sidechain to verify local model updates without exposing sensitive data. A built-in challenge mechanism efficiently detects adversarial behavior. In experiments on image classification and language modeling tasks, ZK-HybridFL achieves faster convergence, higher accuracy, lower perplexity, and reduced latency compared to Blade-FL and ChainFL. It remains robust against substantial fractions of adversarial and idle nodes, supports sub-second on-chain verification with efficient gas usage, and prevents invalid updates and orphanage-style attacks. This makes ZK-HybridFL a scalable and secure solution for decentralized FL across diverse environments.

Figures (16)

• Figure 1: Evolution of a DAG ledger over two consecutive epochs. Nodes are color-coded to indicate their status: red for tips, yellow for unconfirmed, and green for confirmed. The blue text represents the AW associated with each node.
• Figure 2: Full nodes (FNs) maintain a complete DAG (FN_DAG) and sidechain (FN_SC) while handling oracle functions. Light nodes (LNs) use a trimmed DAG (LN_DAG) and lightweight sidechain (LN_SC), with LN_M serving as the LN's core controlling module.
• Figure 3: ZK-HybridFL workflow (1–12): 1) aggregate IDs; 2) fetch blocks; 3) local train; 4) bundle proofs; 5) submit bundle; 6) committee admit; 7) fetch ZKP tips; 8) validate ZKPs; 9) parent selection; 10) attach block; 11) update weights; 12) aggregate and reward. Challenge loop (i–iv): (i) fetch proof; (ii) local GRA; (iii) submit proof; (iv) reward or slash.
• Figure 4: Training loss of Blade-FL, ChainFL, and ZK-HybridFL for $n=15$, $\mu=20\%$ adversaries, $\gamma=10\%$ lazy nodes, $R=5$, and $B=50$.
• Figure 5: Number of detected invalid models over training epochs, corresponding to \ref{['training_loss']}.

Theorems & Definitions (8)

• Remark 2.1
• Remark 2.2
• Remark 2.3
• Remark 4.1
• Remark 4.2
• Remark S1.1
• Lemma S1.2: Bounded-stake collusion
• Remark S1.3