ARC: A Generalist Graph Anomaly Detector with In-Context Learning
Authors: Yixin Liu, Shiyuan Li, Yu Zheng, Qingfeng Chen, Chengqi Zhang, Shirui Pan
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on multiple benchmark datasets from various domains demonstrate the superior anomaly detection performance, efficiency, and generalizability of ARC. |
| Researcher Affiliation | Academia | Yixin Liu1, , Shiyuan Li2, , Yu Zheng3, , Qingfeng Chen2, , Chengqi Zhang4, Shirui Pan1, 1Griffith University, 2Guangxi University, 3La Trobe University, 4The Hong Kong Polytechnic University |
| Pseudocode | Yes | The algorithmic description of the feature alignment in ARC, the training process of ARC, and inference process of ARC are summarized in Algo. 1, Algo. 2, and Algo. 3, respectively. |
| Open Source Code | Yes | The source code of ARC is available at https://github.com/yixinliu233/ARC. |
| Open Datasets | Yes | Specifically, the training datasets Ttrain comprise Pub Med, Flickr, Questions, and Yelp Chi, while the testing datasets Ttest consist of Cora, Cite Seer, ACM, Blog Catalog, Facebook, Weibo, Reddit, and Amazon. For detailed information, please refer to Appendix F.1. |
| Dataset Splits | No | The paper specifies training and testing datasets (Ttrain and Ttest) but does not explicitly define a separate validation set split (e.g., by percentage or count) for its own model training or hyperparameter tuning within the paper text, although it mentions using random search for hyperparameters. |
| Hardware Specification | Yes | All experiments were performed on a Linux server with an Inter Xeon microprocessor E-2288G CPU and a Quadro RTX 6000 GPU. |
| Software Dependencies | Yes | We implemented the proposed ARC using Py Torch 2.1.2, Py Torch Geometric (Py G) 2.3.1, and DGL 0.9.0. |
| Experiment Setup | Yes | We select some key hyper-parameters of ARC through random search within specified grids. Specifically, the random search was performed within the following search space: Hidden layer dimension: {64, 128, 256, 512, 1024} Number of MLP layers: {1, 2, 3, 4} Propagation iteration: {1, 2, 3, 4, 5} Dropout rate: {0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8} Learning rate: floats between 10 5 and 10 2 Weight decay: floats between 10 6 and 10 3 |