Graph Neural Convection-Diffusion with Heterophily
Authors: Kai Zhao, Qiyu Kang, Yang Song, Rui She, Sijie Wang, Wee Peng Tay
IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments, which suggest that our framework can achieve competitive performance on node classification tasks for heterophilic graphs, compared to the state-of-the-art methods. |
| Researcher Affiliation | Collaboration | Kai Zhao1 , Qiyu Kang1 , Yang Song2 , Rui She1 , Sijie Wang1 and Wee Peng Tay1 1Nanyang Technological University 2C3.AI |
| Pseudocode | Yes | Algorithm 1 Neural CDE Inference |
| Open Source Code | Yes | The code is available at https://github.com/zknus/Graph-Diffusion-CDE. |
| Open Datasets | Yes | The paper [Pei et al., 2019] evaluates the performance of their model on six heterophilic graph datasets: Squirrel, Chameleon, Actor, Texas, Cornell, and Wisconsin. ... The datasets Cornell, Texas, and Wisconsin 1 do not have this data leakage issue but are relatively small and have significantly imbalanced classes... 1Available in http://www.cs.cmu.edu/afs/cs.cmu.edu/project/theo11/www/wwkb ... we additionally include six new heterophilic datasets proposed in [Platonov et al., 2023]. |
| Dataset Splits | Yes | For these heterophilic datasets, we follow the data splitting in [Platonov et al., 2023], which is 50%, 25%, and 25% for training, validation, and testing. |
| Hardware Specification | Yes | OOM refers to out-of-memory on NVIDIA RTX A5000 GPU. |
| Software Dependencies | No | The paper mentions using the Adam optimizer [Kingma and Ba, 2014] and solving the neural PDE through methods in [Chen et al., 2018], but does not provide specific version numbers for software dependencies like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | For all these datasets, we use the Adam optimizer [Kingma and Ba, 2014] with a learning rate of 0.01 and weight decay of 0.001. We also apply a dropout rate of 0.2 to prevent overfitting issues. |