GNN-VPA：一种面向图神经网络的方差保持聚合策略
GNN-VPA: A Variance-Preserving Aggregation Strategy for Graph Neural Networks
AI总结：
该研究针对图神经网络表达能力依赖聚合与读出函数的问题，基于信号传播理论提出方差保持聚合策略VPA，在保持表达力的同时优化传播动力学，提升了主流GNN架构的预测性能，为无归一化/自归一化GNN发展提供了方向。
摘要：
图神经网络（GNNs），尤其是消息传递神经网络，在物理、药物发现和分子建模等多个领域表现出色。GNNs区分非同构图的表达能力，关键取决于用于消息聚合和图级读出的函数。通过应用信号传播理论，我们提出了一种方差保持聚合函数（VPA），它在保持表达能力的同时，还能改善前向和反向传播动力学。实验表明，VPA能提升主流GNN架构的预测性能，还能优化学习动力学。我们的结果可为无归一化或自归一化GNNs的发展铺路。
英文摘要：
Graph neural networks (GNNs), and especially message-passing neural networks, excel in various domains such as physics, drug discovery, and molecular modeling. The expressivity of GNNs with respect to their ability to discriminate non-isomorphic graphs critically depends on the functions employed for message aggregation and graph-level readout. By applying signal propagation theory, we propose a variance-preserving aggregation function (VPA) that maintains expressivity, but yields improved forward and backward dynamics. Experiments demonstrate that VPA leads to increased predictive performance for popular GNN architectures as well as improved learning dynamics. Our results could pave the way towards normalizer-free or self-normalizing GNNs.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
