Layer-diverse Negative Sampling for Graph Neural Networks

arXiv:2403.11408v1 Announce Type: new
Abstract: Graph neural networks (GNNs) are a powerful solution for various structure learning applications due to their strong representation capabilities for graph data. However, traditional GNNs, relying on message-passing mechanisms that gather information exclusively from first-order neighbours (known as positive samples), can lead to issues such as over-smoothing and over-squashing. To mitigate these issues, we propose a layer-diverse negative sampling method for message-passing propagation. This method employs a sampling matrix within a determinantal point process, which …

abstract applications arxiv capabilities cs.lg data diverse gather gnns graph graph data graph neural networks however information layer negative networks neural networks positive representation samples sampling solution type