A curated list of resources for OOD detection with graph data.
If some related papers are missing, please contact us via pull requests :)
Unlike traditional anomaly detection, which assumes a small percentage of anomalies, in the graph domain, the OOD part may contain nodes that are comparable in size to the ID part.
(taken from "Learning on Graphs with Out-of-Distribution Nodes", KDD 2022)
Anomaly graphs are usually malicious samples from a real system and available during training; while OOD graphs are samples from the distribution that the model has not seen during training.
(taken from "A Data-centric Framework to Endow Graph Neural Networks with Out-Of-Distribution Detection Ability", KDD 2023)
KDD 2023. A Data-centric Framework to Endow Graph Neural Networks with Out-Of-Distribution Detection Ability. [paper] [code]
- Graph-level, post-hoc, training without outlier exposure.
WSDM 2023. GOOD-D: On Unsupervised Graph Out-Of-Distribution Detection [paper] [code]
- Graph-level, Self-supervised contrastive learning.
🔥 NeurIPS 2022. GraphDE: A Generative Framework for Debiased Learning and Out-of-Distribution Detection on Graphs. [paper] [code]
- Graph-level, Generative, Variational inference.
ICML workshop 2022. Towards OOD Detection in Graph Classification from Uncertainty Estimation Perspective. [paper]
- Graph-level, Uncertainty estimation.
🔥 ICLR 2023. Energy-based out-of-distribution detection for graph neural networks. [paper] [code]
- Node-level with two disconnected graphs, Energy-based score with pagerank-based propagation.
KDD 2022. Learning on Graphs with Out-of-Distribution Nodes. [paper] [code]
- Node-level with inter-edges, Semi-supervised outlier detection, a modified GAT with three regularization terms.
NeurIPS 2021. Graph Posterior Network: Bayesian Predictive Uncertainty for Node Classification. [paper]
- Node-level, Bayesian posterior.
NeurIPS 2020. Uncertainty Aware Semi-Supervised Learning on Graph Data. [paper] [code]
- Node-level, a Graph-based Kernel Dirichlet distribution Estimation (GKDE) method.
NeurIPS 2023. GADBench: Revisiting and Benchmarking Supervised Graph Anomaly Detection. [paper] [code]
ICLR 2023 submission. Revisiting Uncertainty Estimation for Node Classification: New Benchmark and Insights [paper]
NeurIPS 2022. Dual-discriminative Graph Neural Network for Imbalanced Graph-level Anomaly Detection. [paper] [code]
🔥 ICML 2022. Rethinking Graph Neural Networks for Anomaly Detection. [paper] [code]
- Anomalies leads to the ‘right-shift’ of spectral energy.
IJCAI 2022. Raising the Bar in Graph-level Anomaly Detection. [paper] [code]
WSDM 2022. Deep Graph-level Anomaly Detection by Glocal Knowledge Distillation. [paper] [code]
- Graph-level Anomaly Detection, Joint random distillation of graph and node representations with two GNNs (GLocalKD).
Big Data 2021. On Using Classification Datasets to Evaluate Graph Outlier Detection: Peculiar Observations and New Insights. [paper]
- Graph-level, Propagation-based outlier detection method (OCGIN).
- Arxiv 2022. Out-Of-Distribution Generalization on Graphs: A Survey. paper
- THUMNLab/awesome-graph-ood
- Out-Of-Distribution Generalization on Graphs: Paper List