New graph dictionary learning framework uses optimal transport for improved classification

Researchers have developed a new graph dictionary learning (GDL) framework that represents graphs as zero-mean Gaussian distributions derived from their filtered Laplacian. This framework approximates observed graphs using a barycenter of learned atom graphs, measured by a novel filter graph distance (fGOT) metric. The reconstruction error is minimized using a tractable approximation called surrogate fGOT (sfGOT), which is optimized end-to-end via backpropagation and linked to the Hilbert-Schmidt Independence Criterion for maximizing statistical dependence between node spectral embeddings. Experiments show this approach achieves competitive performance in graph clustering and classification tasks. AI

IMPACT Introduces a new method for graph representation and analysis, potentially improving performance in graph-based machine learning tasks.

RANK_REASON The cluster contains a research paper detailing a novel framework for graph dictionary learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• fGOT
• graph classification
• graph dictionary learning
• Hilbert-Schmidt Independence Criterion
• Hugging Face
• Laplace operator
• normal distribution
• sfGOT

• paper
• other

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