NICE framework offers scale-stable explanations for Graph Neural Networks

Researchers have introduced NICE, a novel framework for generating explanations for Graph Neural Networks (GNNs). NICE addresses the issue of 'Scale Drift,' a phenomenon where standard perturbation methods like Element-wise Masking cause a deterministic contraction in message-passing scales, leading to unreliable predictions. NICE utilizes Noise Corruption (NC) as a scale-stable alternative, preserving expected message norms. The framework also incorporates a Stochastic Restoration Boundary (SRB) and Boundary-Integrated Gradient (BIG) to produce more faithful and compact edge attributions. AI

IMPACT Introduces a more reliable method for understanding GNN decision-making, potentially improving trust and debugging in AI applications that use graph structures.

RANK_REASON The cluster contains a research paper detailing a new method for explaining Graph Neural Networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Boundary-Integrated Gradient
• Element-wise Masking
• graph neural network
• NICE
• Perturb-Query
• Scale Drift
• Stochastic Restoration Boundary

• paper
• model release

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