FedHGPrompt: Privacy-Preserving Federated Prompt Learning for Few-Shot Heterogeneous Graph Learning.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Learning from heterogeneous graphs faces challenges with data scarcity and privacy.
• Federated learning and graph prompt learning offer solutions but are difficult to integrate for complex graphs.

Purpose of the Study:

• To propose FedHGPrompt, a novel federated framework for heterogeneous graph learning.
• To address data scarcity and privacy concerns in decentralized graph learning.

Main Methods:

• A three-layer architecture: unification, adaptation, and privacy layers.
• Dual templates for graph/task standardization and trainable dual prompts for few-shot adaptation.
• Cryptographic secure aggregation protocol for robust data privacy.

Main Results:

• FedHGPrompt outperforms existing federated graph learning baselines on real-world datasets (ACM, DBLP, Freebase).
• Achieves superior few-shot learning performance with strong privacy guarantees.
• Demonstrates practical communication efficiency.

Conclusions:

• FedHGPrompt offers an effective approach for collaborative learning on distributed, privacy-sensitive heterogeneous graphs.
• The framework successfully integrates few-shot learning and federated learning for complex graph structures.