Provably Powerful Graph Neural Networks for Directed Multigraphs
Authors: Béni Egressy, Luc von Niederhäusern, Jovan Blanuša, Erik Altman, Roger Wattenhofer, Kubilay Atasu
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To validate the effectiveness of our proposed adaptations in practice, we conduct experiments on synthetic subgraph detection tasks, which demonstrate outstanding performance with almost perfect results. Moreover, we apply our proposed adaptations to two financial crime analysis tasks. |
| Researcher Affiliation | Collaboration | 1ETH Zurich, Zurich, Switzerland 2IBM Research Europe, Zurich, Switzerland 3IBM Watson Research, Yorktown Heights, NY, USA |
| Pseudocode | Yes | Details of the generator and pseudocode can be found in the appendix. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. It links to the arXiv version of the paper, not a code repository. |
| Open Datasets | Yes | We use simulated money laundering data (Altman et al. 2023)." "We use an Ethereum transaction network published on Kaggle (Chen et al. 2021), where some nodes are labeled as phishing accounts." "Chameleon, Squirrel (Pei et al. 2020), and Arxiv-Year (Hu et al. 2020) |
| Dataset Splits | Yes | We use a 60-20-20 temporal train-validation-test split, i.e., we split the transactions after ordering them by their timestamps. [...] We use a temporal train-validation-test split, but this time splitting the nodes. We use a 65-15-20 split because the illicit accounts are skewed towards the end of the dataset. |
| Hardware Specification | No | The paper only mentions 'on a single GPU' for inference rate, but does not provide specific hardware details such as GPU model, CPU model, or memory for running experiments. |
| Software Dependencies | No | The paper mentions software components like GIN, GAT, PNA, XGBoost, and Light GBM, but does not specify their version numbers or other ancillary software with versions. |
| Experiment Setup | No | The paper states 'Further details of the experimental setup for the different datasets can be found in the appendix,' indicating that these details are not present in the main text. |