PoCQ consensus boosts federated learning accuracy by 34%

A team of researchers from Australia and Bangladesh has unveiled Proof of Contribution Quality (PoCQ), a blockchain-based consensus framework designed to secure decentralized federated learning (FL) against model poisoning attacks. Traditional decentralized FL removes the central coordinator but introduces vulnerabilities—malicious clients can submit poisoned updates, and validation overhead grows quickly. PoCQ solves both problems with a reputation-driven mechanism that evaluates client contributions using cryptographic commitments and lightweight norm-based checks. Only compact audit metadata is stored on the blockchain, preserving scalability while enabling efficient detection of bad actors.

Experiments across three benchmark datasets (including challenging non-iid medical data) showed PoCQ outperforming all prior state-of-the-art methods. It achieved accuracy gains of 34.1% on difficult medical scenarios and an 11% improvement in global average accuracy. Additionally, PoCQ reduced validation time per round by 21.27% on average, proving that robust security doesn't have to come at the cost of efficiency. The framework is particularly promising for healthcare, finance, and other domains where data privacy and trust are paramount.