EVENT DOSSIER
On September 7, 2026, the arXiv cs.AI journal published a research paper titled “Global to Local: Topology-Preserving Adaptive Graph Pooling via Granular-Ball”. This study proposes a new graph neural network technique aimed at achieving adaptive graph pooling from global to local scales using granular balls. The core advantage of this method is its ability to maintain the topological structure of graphs while adapting to data aggregation requirements at different scales, providing a new solution for processing large-scale non-Euclidean data.
arXiv:2609.04978v1 提出一种基于细粒度球体的拓扑保持自适应图池化（TPAGP）方法，旨在解决现有工作忽视全局到局部模式及自适应粒度的问题。该方法通过整合节点特征与拓扑信息动态将图划分为细粒度球，生成多粒度表示以捕捉局部和全局结构模式，并设计了跨粒度交互优化的多粒度图网络模型。实验表明，TPAGP 在多个基准数据集上优于现有池化方法，有效缓解了固定粒度策略导致的信息丢失。