GNN论文粗读

论文

随缘更新

基于异构图的GNN论文

1. Distance Information Improves Heterogeneous Graph Neural Networks:DOI: 10.1109/TKDE.2023.3300879
   • 转导和归纳任务，创新点：异构距离编码HDE提高GNN表现能力
2. Heterogeneous Graph Neural Network via Attribute Completion:https://dl.acm.org/doi/10.1145/3442381.3449914
   • 属性补全任务，创新点：包括拓扑嵌入的预学习和基于注意力机制的属性补全
3. Composite Graph Neural Networks for Molecular Property Prediction：doi:10.3390/ijms25126583
   • 分类和回归任务，创新点：复合图神经网络使用多个状态更新网络处理异构图，每个网络专用于特定节点类型
   • 该文章具有回归任务，提供了使用基于异构图的GNN进行数值预测的理论与实践证明
4. Equivariant Line Graph Neural Network for Protein-Ligand Binding Affinity Prediction：doi:10.1109/JBHI.2024.3383245
   • 结合亲和力预测（数值预测），创新点：3d坐标复合体
5. Improving airport arrival flow prediction considering heterogeneous and dynamic network dependencies：https://doi.org/10.1016/j.inffus.2023.101924
   • 预测机场到达量，创新点：动态多图神经网络（卷积+注意力），时间感知注意力，重校准融合模块
6. Estimating package arrival time via heterogeneous hypergraph neural network：doi:10.1016/j.eswa.2023.121740
   • 预测到达时间，创新点：利用超图解决ETA预测问题
7. A city-based PM2.5 forecasting framework using Spatially Attentive Cluster-based Graph Neural Network model：doi:10.1016/j.jclepro.2023.137036
   • 短期PM2.5浓度，创新点：建模程序考虑了相关的气象变量
8. Sequence pre-training-based graph neural network for predicting lncRNA-miRNA associations：doi:10.1093/bib/bbad317
   • 二分类任务

GNN领域论文

1. Denoising AggrDegation of Graph Neural Networks by Using Principal Component Analysis：doi:10.1109/TII.2022.3156658
   • 去噪任务，创新点：dropout+PCA降低运算成本
2. Accelerating Distributed GNN Training by Codes：doi:10.1109/TPDS.2023.3295184
   • 一般GNN任务，创新点：引入编码技术降低GNN的通信开销
3. Dual-stream GNN fusion network for hyperspectral classification：doi:10.1007/s10489-023-04960-3
   • 分类任务，创新点：使用子立方体作为输入降低计算成本，应用了图池化、局部引导模块
4. SCV-GNN: Sparse Compressed Vector-based Graph Neural Network Aggregation：doi:10.1109/TCAD.2023.3291672
   • 一般GNN任务，创新点：针对聚合操作优化数据结构，使用Z-Morton排序推到基于数据局部性的计算排序和分区方案
5. Edgeless-GNN: Unsupervised Representation Learning for Edgeless Nodes：doi:10.1109/TETC.2023.3292240
   • 新节点嵌入
6. Ha-gnn: a novel graph neural network based on hyperbolic attention：doi:10.1007/s00521-024-09689-9
   • 分类任务，创新点：将图结构映射至双曲空间或者其切线空间（HGNN），HGNN+注意力$\to$​HA-GNN
7. Fast prediction and control of air core in hydrocyclone by machine learning to stabilize operations：doi:10.1016/j.jece.2023.111699
   • 再现空气剖面？创新点：CFD+GNN，数据平滑，损失函数调整以纳入CFD的空气核心信息；将GNN与随机森林结合；将模型与遗传算法结合

环境领域GNN论文

1. Urban wind field prediction based on sparse sensors and physics-informed graph-assisted auto-encoder：doi:10.1111/mice.13147
   • 风场方向预测，创新点：PINN+GNN+编码解码
   • 不做考虑
2. A two-stage CFD-GNN approach for efficient steady-state prediction of urban airflow and airborne contaminant dispersion：doi:10.1016/j.scs.2024.105607

• 风、阵风、污染物扩散预测，创新点：CFD提供初始状态，gnn进行后续推理从而加速运算；使用具有$k-\epsilon $模型的SRANS为GNN提供信息丰富的初始状态。

3. A new integrated prediction method of river level based on spatiotemporal correlation：doi:10.1007/s00477-023-02617-8

   • 河流水位预测，创新点：
     1. pearson相关性分析，建立时间相关模型
     2. ChebNet（GNN）
     3. 利用AE-XGBoost重建时间特征并进行预测

4. Nationwide Air Pollution Forecasting with Heterogeneous Graph Neural Networks：doi:10.1145/3637492

   • 空气污染预测，创新点：寄，摘要没写，代码没有

5. A long-term prediction method for PM2.5 concentration based on spatiotemporal graph attention recurrent neural network and grey wolf optimization algorithm：doi:10.1016/j.jece.2023.111716

   • PM2.5，创新点：GWO，GAT、GNN、GRU$\to$GART

6. PM2.5 forecasting under distribution shift: A graph learning approach：https://doi.org/10.1016/j.aiopen.2023.11.001

   • https://github.com/yachuan/pm2.5forecasting
   • https://github.com/yachuan/pm2.5forecasting