Deep Graph Mating

Neural Information Processing Systems 

We strive to create a child Graph Neural Network (GNN) that integrates knowledge from pretrained parent models without requiring re-training, fine-tuning, or annotated labels.