Directional diffusion models for graph representation learning
Authors: Run Yang, Yuling Yang, Fan Zhou, Qiang Sun
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In order to assess the effectiveness of our proposed models, we conduct extensive experiments on 12 publicly available datasets, with a particular focus on two distinct graph representation learning tasks. The experimental results unequivocally establish the superiority of our models over state-of-the-art baselines, underscoring their effectiveness in capturing meaningful graph representations. Our code is available at https://github.com/statsle/DDM. Numerically, we perform experiments on 12 benchmark datasets covering both node and graph classification tasks. The results consistently highlight the superior performance of our models when compared to state-of-the-art contrastive learning and generative approaches (Hou et al., 2022). Additionally, we provide comprehensive ablation studies to gain a deeper understanding of the mechanisms underlying directional diffusion models. |
| Researcher Affiliation | Collaboration | Run Yang SUFE and Baidu luckyyangrun@163.sufe.edu.cn; Yuling Yang SUFE sibyllayang@163.sufe.edu.cn; Fan Zhou SUFE zhoufan@mail.shufe.edu.cn; Qiang Sun University of Toronto qiang.sun@utoronto.ca |
| Pseudocode | No | The paper describes the algorithm in text and mentions that the 'complete pipeline is presented in the appendix', but it does not include a clearly labeled pseudocode or algorithm block in the main body. |
| Open Source Code | Yes | Our code is available at https://github.com/statsle/DDM. |
| Open Datasets | Yes | We conduct extensive experiments on 12 publicly available datasets, with a particular focus on two distinct graph representation learning tasks. We perform experiments on 12 benchmark datasets covering both node and graph classification tasks...These experiments are conducted on seven widely-used datasets, namely MUTAG, IMDB-B, IMDB-M, PROTEINS, COLLAB, and REDDIT-B (Yanardag and Vishwanathan, 2015). To assess the quality of the node-level representations produced by our method, we conduct evaluations of DDM on six standard benchmark datasets: Cora, Citeseer, Pub Med (Yang et al., 2016), Ogbn-arxiv (Hu et al., 2020a), Amazon-Computer (Zhang et al., 2021), and Amazon-Photo (Zhang et al., 2021). |
| Dataset Splits | No | The paper mentions training, and testing, and evaluation protocols, but it does not explicitly provide specific details about training/validation/test dataset splits (e.g., percentages or sample counts) needed to reproduce the experiment, nor does it cite specific predefined splits for all datasets. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments. It only mentions experiments were conducted. |
| Software Dependencies | No | The paper mentions using LIBSVM but does not provide version numbers for any software dependencies. It also refers to PyTorch (indirectly through citations of other papers that might use it), but no specific version for its own implementation. |
| Experiment Setup | Yes | Then, we extract feature representations from diffusion steps 50, 100, 200 using the pre-trained model. Additional details on hyper-parameters can be found in the appendix. Additional details regarding hyperparameters can be found in the appendix. |