New benchmark Lethe evaluates edge unlearning in graph neural networks

Researchers have introduced Lethe, a new benchmark designed to evaluate edge unlearning methods for graph neural networks. Current evaluations for edge unlearning are limited, often using datasets where node features overshadow graph structure and employing accuracy metrics that are insensitive to edge removal. Lethe addresses these issues by focusing on large graphs, offering tasks of varying difficulty, and utilizing link inference attacks as a more effective evaluation measure. The benchmark includes 15 methods and 8 datasets, aiming to provide a more robust and reproducible framework for assessing edge unlearning techniques. AI

IMPACT Provides a standardized evaluation framework for edge unlearning, potentially improving the robustness and efficiency of data removal techniques in graph neural networks.

RANK_REASON The item is a research paper introducing a new benchmark for evaluating specific machine learning methods. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• CiteSeerX
• Cora
• Edge Unlearning
• graph neural networks
• Graph Unlearning with Efficient Partial Retraining
• Lethe
• link inference attacks

• paper
• other

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