New UGPrompt framework enables unsupervised GNN adaptation without source data

Researchers have introduced UGPrompt, a novel framework for unsupervised adaptation of Graph Neural Networks (GNNs). This method addresses the challenge of adapting GNNs to new tasks without access to labeled data or the original training data, while keeping the pre-trained GNN entirely frozen. UGPrompt employs consistency regularization and pseudo-labeling to train a prompting function, incorporating diversity and domain regularization to handle distribution mismatches and class imbalances. Experiments show that UGPrompt surpasses current supervised prompting techniques, even when those methods have access to labeled data, highlighting the potential of unsupervised prompting for practical GNN adaptation. AI

IMPACT This research could enable more efficient and accessible adaptation of GNNs for real-world applications where labeled data is scarce.

RANK_REASON The cluster contains a research paper detailing a new framework for GNN prompting. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Graph Neural Networks
• Peyman Baghershahi
• UGPrompt

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →