From Trainable Negative Depth to Edge Heterophily in Graphs
Authors: Yuchen Yan, Yuzhong Chen, Huiyuan Chen, Minghua Xu, Mahashweta Das, Hao Yang, Hanghang Tong
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate the superiority of TEDGCN on node classification tasks for a variety of homophilic and heterophilic graphs. |
| Researcher Affiliation | Collaboration | 1University of Illinois at Urbana Champaign, IL, USA 2Visa Research, CA, USA |
| Pseudocode | No | The paper describes the proposed algorithms (TEDGCN-S and TEDGCN-D) using mathematical equations and textual explanations, but it does not provide formal pseudocode blocks or algorithms. |
| Open Source Code | No | Since this work is conducted during internship in Visa Research and the code can not be made public according to the policy of Visa Research. If you have any questions about the implementation, please email yucheny5@illinois.edu to get some help. |
| Open Datasets | Yes | We use 11 datasets for evaluation, including 4 homophilic graphs: Cora [28], Citeseer [28], Pubmed [28] and DBLP [6], and 7 heterophilic graphs: Cornell [41], Texas [41], Wisconsin [41], Actor [41], Chameleon [44], Squirrel [44], and cornell5 [15]. We collect all datasets from the public GCN platform Pytorch-Geometric [15]. |
| Dataset Splits | Yes | For Cora, Citeseer, Pubmed with data splits in Pytorch-Geometric, we keep the same training/validation/testing set split as in GCN [28]. For the remaining 8 datasets, we randomly split every dataset into 20/20/60% for training, validation, and testing. ... we compare all methods performances on 4 commonly used heterophilic graph datasets: Texas, Cornell, Wisconsin and Actor, under the same public 48%/32%/20% split for training, validation and testing. |
| Hardware Specification | Yes | All experiments are run on a Tesla-V100 GPU. |
| Software Dependencies | No | The paper mentions collecting datasets from 'Pytorch-Geometric [15]' but does not provide specific version numbers for Pytorch-Geometric or any other software dependencies. |
| Experiment Setup | Yes | We set the learning rate to 0.005 or 0.01, the decaying weight for the learning rate to 5e 4, the number of training epochs to 500. The K largest/smallest eigenvalues for TEDGCN-D is max{1000, 0.1n}, where n is the number of nodes in the input graph. |