XinyiYS/CollaborativeFairFederatedLearning

This project is a classic research artifact, serving as the official implementation for the paper 'Collaborative Fairness in Federated Learning' (likely Lyu et al.). With 55 stars and zero velocity over a 6-year period, it functions as a historical reference rather than a living tool. The defensibility is very low because the core logic—adjusting global model distribution based on a 'contribution' metric—is an algorithmic pattern that has since been integrated into or surpassed by production-grade Federated Learning (FL) frameworks. From a competitive standpoint, any organization building a serious FL system would use modern frameworks like Flower (flwr.dev), NVIDIA FLARE, or OpenMined's PySyft. These platforms either already have 'fairness' plugins or make it trivial to implement the logic described in this repo. The 'Collaborative Fairness' concept is highly relevant for B2B consortiums (e.g., banks or hospitals sharing data), but this specific codebase lacks the security, communication protocols, and scalability required for those environments. Frontier labs face low risk here because they focus on large-scale centralized pre-training or private aggregation (DP), whereas this is a niche 'incentive' layer for decentralized coordination. It is highly likely to be entirely displaced by modern FL library modules within a very short horizon if it hasn't been already.