PoCQ: Proof of Contribution Quality as a Lightweight Blockchain Consensus for Secure Federated Learning
For decentralized federated learning systems, PoCQ provides a lightweight, secure consensus mechanism that improves robustness against poisoning attacks while maintaining efficiency.
PoCQ introduces a blockchain consensus for decentralized federated learning that uses reputation-aware validation and aggregation to detect malicious updates efficiently. It achieves up to 34.1% accuracy gains on medical datasets and reduces validation time by 21.27% per round.
Decentralized Federated Learning (FL) removes reliance on centralized coordinators but remains vulnerable to model poisoning, unreliable validation, and high validation overhead. This paper introduces Proof of Contribution Quality (PoCQ), a blockchain-based consensus framework designed to secure decentralized FL through reputation-aware validation and aggregation. PoCQ evaluates client updates using cryptographic commitments and lightweight norm-based validation, enabling efficient detection of malicious contributions while limiting validation cost. A reputation-driven consensus mechanism dynamically adjusts the influence of participants based on their historical contribution quality, while the blockchain stores only compact audit metadata to preserve scalability. Extensive experiments under poisoning scenarios across three benchmark datasets demonstrate that PoCQ outperforms the strongest state-of-the-art methods, achieving accuracy gains of 34.1% on challenging medical datasets in highly non-iid settings and an 11% improvement in global average accuracy. In addition, PoCQ reduces validation time by 21.27% on average per round, highlighting its effectiveness in jointly enhancing robustness and efficiency for fully decentralized federated learning.