DeGLIF technique enhances GNN node classification by denoising noisy graph data

Researchers have developed DeGLIF, a novel technique designed to improve node classification accuracy in Graph Neural Networks (GNNs) by addressing label noise in datasets. This method utilizes a leave-one-out influence function to approximate the impact of removing individual training points on model parameters and validation loss. DeGLIF employs a theoretically motivated relabeling function to denoise the training data without requiring prior knowledge of the noise model or level. Experimental results demonstrate that DeGLIF outperforms existing baseline algorithms in accuracy. AI

IMPACT This method could improve the reliability of graph-based AI models trained on real-world, potentially noisy, datasets.

RANK_REASON The cluster contains an arXiv preprint detailing a new method for graph neural networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• DeGLIF
• Graph Neural Networks

• paper
• model release

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