New method accelerates temporal GNN training with adaptive pseudo-supervision

Researchers have developed a new method called Moving-Averaged Labels (MAL) to improve the training of temporal graph neural networks (GNNs). This technique addresses the issue of irregular supervision in real-world dynamic graphs by assigning soft pseudo-targets based on past supervised signals. MAL aims to reduce gradient variance and accelerate convergence without altering the model architecture or loss function. Experiments show that MAL significantly boosts predictive performance and achieves faster time-to-accuracy, establishing a new state-of-the-art on common Temporal Graph Benchmark datasets. AI

IMPACT Accelerates training for temporal graph neural networks, potentially enabling more efficient real-world dynamic graph analysis.

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

Read on arXiv cs.LG →

• Alexander Panyshev
• DyRepv2
• Moving-Averaged Labels
• Temporal Graph Benchmark
• temporal graph neural networks
• TGNv2

• paper
• infra

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