Highly Imperceptible Black-Box Graph Injection Attacks with Reinforcement Learning
Authors: Maochang Zhao, Jing Zhang
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on real-world datasets and mainstream GNN models demonstrate that the proposed TEANI poses more effective and imperceptible threats than state-of-the-art attack methods. |
| Researcher Affiliation | Academia | Maochang Zhao, Jing Zhang* School of Cyber Science and Engineering, Southeast University, Nanjing, China EMAIL |
| Pseudocode | Yes | Algorithm 1: The training algorithm of TEANI |
| Open Source Code | No | The paper does not provide any explicit statement about releasing source code, a link to a repository, or mention of code in supplementary materials. |
| Open Datasets | Yes | Following previous works (Ju et al. 2023; Liu, Huang, and Zhao 2024; Sun et al. 2020), we conducted experiments on three public datasets: Cora, Cora-ML (Mccallum et al. 2000), and Citeseer (Sen et al. 2008). |
| Dataset Splits | No | The paper states: "The nodes are divided into a training set VL and a test set VU, where VL VU = V . In the node classification task, only the labels of the training set VL are accessible and used to train the GNNs model, while the labels of the test set VU are used to evaluate the model s predictive performance." However, it does not specify exact percentages, sample counts, or specific methods for generating these splits for the datasets used (Cora, Citeseer, Cora-ML). |
| Hardware Specification | Yes | All experiments are conducted on an Ubuntu server equipped with an Intel Xeon 8336C CPU and an RTX 4080 GPU. |
| Software Dependencies | No | The paper mentions algorithms and models like PPO, GCL, GNNs, and GCN, but does not specify any software libraries or programming languages with their version numbers that were used for implementation. |
| Experiment Setup | Yes | The experimental setup for TEANI is as follows: the state embedding network consists of two layers of GCL with a hidden dimension of 128. In the PPO, the learning rate is set to 10 4, the minibatch size is 30, and the clipping coefficient ϵ is set at 0.1. In GAE, the discount factors γ and parameter λ are 0.95 and 0.99, respectively. |