Knowledge-Free Correlated Agreement for Incentivizing Federated Learning

TL;DR

The paper proposes KFCA, a novel reward mechanism for federated learning that incentivizes honest client contributions without needing ground truth or distribution info, and is suitable for decentralized systems.

Contribution

It introduces Knowledge-Free Correlated Agreement (KFCA), addressing label-flipping vulnerabilities and enabling real-time, decentralized reward computation in federated learning.

Findings

KFCA is strictly truthful under honest majority.

KFCA performs well in federated LLM adapter tuning.

KFCA enables efficient real-time rewards in blockchain settings.

Abstract

We introduce Knowledge-Free Correlated Agreement (KFCA) to reward client contributions in federated learning (FL) without relying on ground truth, a public test set, or distribution knowledge. Under categorical reports and an honest majority, KFCA is strictly truthful, addressing the label-flipping vulnerability of Correlated Agreement (CA). We evaluate KFCA on federated LLM adapter tuning and a real-world PCB inspection task, showing efficient real-time reward computation suitable for decentralized and blockchain-based incentive designs.