本仓库整理了我们综述论文 '当图神经网络遇到因果性：机遇、方法和展望' 中审查的因果性启发式图神经网络（CIGNN）工作及相关资源的精选列表。最近，将因果学习技术集成到 GNN 中展示了在缓解可信性问题方面的重要潜力。这通过捕捉数据背后的因果性而非依赖于表面的相关性来实现。在本文综述中，我们从因果性的视角全面回顾了 CIGNNs 的最新进展，旨在提炼这些工作中背后的基本原理，并激发对该有前途方向的进一步研究。
-
(Arxiv 2202.07987) H. Li, et al., “图上的离分布泛化：一个综述,” CoRR, 卷 abs/2202.07987, 2022. [论文] -
(NIPS'22) GOOD：一个图的分布外基准，[论文], [Github] -
(Arxiv 2201.09637) Drugood：分布外（OOD）数据集策展人和辅助药物发现的人工智能基准 - 重点关注带噪声标注的亲和力预测问题。[论文], [Github] -
(Arxiv 2310.03152) 关于化学动力学性质分布外泛化预测的研究进展。[论文], [Github]
(Arxiv 2204.09888) 图数据挖掘中的公平性：一项综述。[Paper], [Github]
(KDD'24) 解决公平图学习数据集中的不足：迈向新的基准。[Paper], [Github]
(NIPS'24) 公平意识图学习的基准。[Paper], [Github]
-
(TPAMI'23) 图网络中的可解释性：一种分类综述。[论文] -
(CSUR'23) 图反事实解释综述：定义、方法、评估和研究挑战。[论文]
-
(NIPS'22) GOOD：一个图的分布外基准，[Paper]，[Github] -
(CIKM'22) GRETEL: 图结构反事实解释评估框架. [Paper], [Github] -
(WSDM'23) 开发和评估基于 GRETEL 的图反事实解释。[论文], [GitHub]
(CIKM'21) Causebox：一个用于基准测试机器学习方法下的处理效应估计器的因果推理工具箱。[论文], [GitHub]
(Arxiv 2307.16405) Causal-learn: Python 中的因果发现。[论文], [Github]
(ICLR'23) 3DIdentBox: 一个可识别性基准测试工具箱。[论文], [Github]
(Arxiv 2201.08802) 图神经网络中的去混杂到解释评估。
(AAAI'23) 面对混杂效应的鲁棒因果图表示学习。
(TNNLS'22) 谨慎标签选择偏差下的去偏差图神经网络。
(TPAMI'23) 在分布外图上的图神经网络泛化.
(TKDE'23) OOD-GNN：分布外广义图神经网络.
(AAAI'24) 学习重新加权以实现可泛化的图神经网络.
(SIGIR'21) 图结构卷积应该信任邻居吗？一种简单的因果推理方法。
(IJCAI'23) 基于因果推理的图神经网络注意力监督：更强大且更简单的选择。
(KDD'23) 基于多样反驳证据的谣言检测。
(UAI'21) 向统一框架公平且稳定的图表示学习迈进。
(ICDM'21) 多视图置信校准框架以实现公平和稳定的图表示学习。
(ICML'21) 图神经网络的生成因果解释。
(TPAMI'23) 图神经网络的强化因果解释器。
(KDD'24) 从重新平衡角度重新思考公平图神经网络。
(ICML'22) 从反事实链接学习以进行链接预测。
(ICDM'23) 通过实际反事实样本缓解图神经网络中的多源偏差。
(WSDM'22) 学习具有图反事实公平性的公平节点表示。
(2023) 图上的反事实公平性：增强学习、隐藏的混杂因素和可识别性。
(AISTAT'22) Cf-gnnexplainer: 图神经网络的反事实解释。
(IJCNN'21) Meg: 生成深度图网络的分子反事实解释。
(NIPS'22) CLEAR: 基于图的生成性反事实解释.
(WWW'22) 基于反事实和事实推理学习和评估图神经网络解释.
(KDD'21) 基于反事实图的可解释脑网络分类。
(NIPS'21) 图上神经网络的鲁棒反事实解释。
(ICDM'21) 多目标解释 GNN 预测。
(WSDM'23) 图神经网络的全局反事实解释器。
(WWW'24) 游戏理论反事实解释图神经网络。
(AAAI'24) 在图神经网络中重新思考因果关系学习。
(KBS'24) 将微小的因果结构引入图表示学习。
- (ICLR'22) Handling distribution shifts on graphs: An invariance perspective.
- (ICLR'22) Discovering invariant rationales for graph neural networks.
- (CVPR'23) Mind the label shift of augmentation-based graph OOD generalization.
- (NIPS'22) Learning invariant graph representations for out-of-distribution generalization.
- (ICDE'23) BA-GNN: on learning bias- aware graph neural network.
- (TOIS'24) Invariant node representation learning under distribution shifts with multiple latent environments.
- (KBS'24) Fortune favors the invariant: Enhancing GNNs' generalizability with Invariant Graph Learning.
- (KDD'22) Causal atten- tion for interpretable and generalizable graph classification.
- (NIPS'22) Debiasing graph neural networks via learning disentangled causal substructure.
- (NIPS'22) Learning substructure invariance for out-of-distribution molecular representations.
- (NIPS'22) Learning causally invariant representa- tions for out-of-distribution generalization on graphs.
- (CVPR'22) Orphicx: A causality-inspired latent variable model for interpreting graph neural networks.
- (KDD'23) Shift-robust molecular relational learning with causal substructure.
- (NIPS'23) Does invariant graph learning via environment augmentation learn invariance?
- (NIPS'23) Joint learning of label and environment causal independence for graph out-of-distribution generalization.
- (CIKM'23) Causality and independence enhancement for biased node classification.
- (CIKM'23) Towards fair graph neural networks via graph counterfactual.
- (AAAI'24) A twist for graph classification: Optimizing causal information flow in graph neural networks.
- (WWW'24) Graph out-of-distribution generalization via causal intervention.
- (ICML'24) Learning Divergence Fields for Shift-Robust Graph Representations.
- (2024) Incorporating Retrieval-based Causal Learning with Information Bottlenecks for Interpretable Graph Neural Network.
- (2024) Unifying invariance and spuriousity for graph out-of- distribution via probability of necessity and sufficiency.
- (2024) CI-GNN: A granger causality- inspired graph neural network for interpretable brain network- based psychiatric diagnosis.
- (2024) Enhancing graph neural networks for self-explainable modeling: A causal perspective with multi-granularity receptive fields.
- (ICML'22) Let invariant rationale discovery inspire graph contrastive learning.
- (KDD'23) FLOOD: A flexible invariant learning framework for out-of-distribution generalization on graphs.
- (NIPS'23) Learning invariant molecular representation in latent discrete space.
- (WWW'23) Generating counterfactual hard negative samples for graph contrastive learning.
- (AAAI'24) Graph contrastive invariant learning from the causal perspective.
- (AAAI'24) Rethinking Dimensional Rationale in Graph Contrastive Learning from Causal Perspective.
- (ICML'24) Disentangled Graph Self-supervised Learning for Out-of-Distribution Generalization.
- (NN'25) Disentangled contrastive learning for fair graph representations.
We appreciate your kind contributions and suggestions! If you know of any causality-inspired GNN works that are not listed, feel free to open an issue. We will update them in our survey and repository.
This project is licensed under the MIT License.
If you find our work useful for your research, please consider citing
@article{jiang2023survey,
title={When Graph Neural Network Meets Causality: Opportunities, Methodologies and An Outlook},
author={Jiang, Wenzhao and Liu, Hao and Xiong, Hui},
journal={arXiv preprint arXiv:2312.12477},
year={2024}
}