Graph neural networks (GNNs), deep learning models specialized for graph understanding, are becoming increasingly popular in research and industry, including for search and recommendation systems. GNNs model networks as a constellation of nodes and edges where nodes connected by one or more edges are considered neighbors. To improve the performance and scalability of GNNs, LinkedIn has OSS’ed PASS (Performance Adaptive Sampling Strategy), a neighborhood sampler that uses an AI model to select the neighbors that will optimize the GNN’s predictive accuracy.