Why Propagate Alone? Parallel Use of Labels and Features on Graphs
Authors: Yangkun Wang, Jiarui Jin, Weinan Zhang, Yang Yongyi, Jiuhai Chen, Quan Gan, Yong Yu, Zheng Zhang, Zengfeng Huang, David Wipf
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Instead, in this section we will focus on conducting experiments that complement our analysis from Section 3 and showcase the broader application scenarios from Section 4. We use four relatively large datasets for evaluation, namely Cora-full, Pubmed (Sen et al., 2008), ogbn-arxiv and ogbn-products (Hu et al., 2020). We report the average classification accuracy and standard deviation after 10 runs with different random seeds, and these are the results on the test dataset when not otherwise specified. |
| Researcher Affiliation | Collaboration | Yangkun Wang1 , Jiarui Jin1 , Weinan Zhang1, Yongyi Yang2 , Jiuhai Chen3 , Quan Gan4, Yong Yu1, Zheng Zhang4, Zengfeng Huang2, David Wipf4 1Shanghai Jiao Tong University, 2Fudan University, 3University of Maryland, 4Amazon |
| Pseudocode | No | The paper does not contain any pseudocode or algorithm blocks. It describes methods and models using mathematical formulations and textual descriptions. |
| Open Source Code | No | The paper does not provide any statement or link indicating the release of source code for the methodology described. |
| Open Datasets | Yes | We use four relatively large datasets for evaluation, namely Cora-full, Pubmed (Sen et al., 2008), ogbn-arxiv and ogbn-products (Hu et al., 2020). |
| Dataset Splits | Yes | For Cora-full and Pubmed, we randomly split the nodes into training, validation, and test datasets with the ratio of 6:2:2 using different random seeds. For ogbn-arxiv and ogbn-products, we adopt the standard split from OGB (Hu et al., 2020). |
| Hardware Specification | No | The paper does not specify any hardware details (e.g., GPU/CPU models, memory, or specific computing platforms) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific version numbers for any software dependencies or libraries used in the experiments. |
| Experiment Setup | Yes | In the implementation, we set λ as 0.6 and use 50 propagation steps. We search the best splitting probability α in the range of {0.05, 0.1, . . . , 0.95}. For TWIRLS, we use 2 propagation steps for the linear propagation layers and λ is set to 1. Our SIGN model also use 5 propagation steps, and we tune the number of MLP layers from 1 or 2 on each dataset. And for SGC, the number of propagation steps is set to 3, and there is one extra linear transformation after the propagation steps. For the linear model, λ is set to 0.9 and the number of propagation steps is 9. The GCN has 256 hidden channels and an activation function of Re LU. We first trained an MLP with exactly the same hyperparameters as in Huang et al. (2021). For each splitting probability α {0.1, 0.2, , 0.9}, we generate 10 splits and precompute Y s and Y c. |