👉 A privacy-preserving user-item graph extension protocol to expand local graphs and convey high-order information while maintaining privacy 🔒

👉 FedPerGNN yields 📉 4.0% – 9.6% reduced errors than state-of-the-art federated customization algorithms under adequate privacy protection, according to experimental results on six datasets for personalization in diverse circumstances.

👉 Furthermore, this method is not restricted to the customization scenario. It may be used as a fundamental strategy for privacy-preserving data mining on decentralized graph data, thus facilitating research in various domains involving graph-structured data.

Continue reading | Check out the paper and github

Source link