Gaussian Perturbations Prevent Oversmoothing in Recurrent GNNs

Researchers have developed a novel method using persistent Gaussian perturbations to combat oversmoothing in recurrent graph neural networks (GNNs). This technique injects independent Gaussian noise after each propagation step, transforming the network into a stochastic dynamical system. Theoretical analysis demonstrates that this approach prevents node representations from collapsing by ensuring a positive lower bound on the expected stationary Dirichlet energy, thereby preserving representation diversity. AI

IMPACT This research offers a new theoretical approach to improving the depth and performance of graph neural networks by addressing the fundamental oversmoothing problem.

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

Read on arXiv cs.LG →

• arXiv
• DagsHub
• Dirichlet energy
• Gaussian function
• graph neural networks
• Hugging Face
• Markov chain
• Mostafa Haghir Chehreghani

• paper
• model release

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