Graph neural network depth value determined by Kesten-Stigum ratio

A new paper explores the optimal depth for graph neural networks on sparse graphs, focusing on node classification within the contextual stochastic block model. The research establishes that the network's performance is dictated by the Kesten-Stigum ratio, a measure related to signal attenuation and average degree. Below a critical threshold, increasing depth yields diminishing returns, while above it, performance improves geometrically, albeit with a theoretical floor. AI

IMPACT This research provides theoretical insights into the optimal depth of graph neural networks for sparse graph analysis, potentially influencing future model architectures.

RANK_REASON The cluster contains a single academic paper detailing theoretical research on graph neural networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• BP
• Cauchy
• CLMP
• Galton–Watson process
• Kesten-Stigum
• Kesten-Stigum ratio
• Poisson
• Poisson Galton-Watson tree

• paper
• other

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