Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Towards Deepening Graph Neural Networks: A GNTK-based Optimization Perspective
Authors: Wei Huang, Yayong Li, weitao Du, Richard Xu, Jie Yin, Ling Chen, Miao Zhang
ICLR 2022 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental evaluation consistently confirms using our proposed method can achieve better results compared to relevant counterparts with both infinite-width and finite-width. |
| Researcher Affiliation | Academia | Wei Huang University of Technology Sydney EMAIL Yayong Li University of Technology Sydney EMAIL Weitao Du Northeastern University EMAIL Jie Yin The University of Sydney EMAIL Richard Yi Da Xu & Ling Chen University of Technology Sydney EMAIL Miao Zhang Aalborg University EMAIL |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions using and adapting existing open-source implementations (GNTK and Drop Edge) but does not provide a specific link or explicit statement for the open-source release of their *own* developed methodology (Critical Drop Edge) or adapted code. |
| Open Datasets | Yes | Details of four real-world graph datasets used for node classification are summarized in Table 3 in Appendix F.1. ... The Cora dataset consists of 2,708 scientific publications... The Citeseer dataset consists of 3,312 scientific publications... The Pubmed Diabetes dataset consists of 19,717 scientific publications... The Physics dataset consists of 34,493 authors as nodes... |
| Dataset Splits | Yes | Table 3: Details of Datasets... Train/Val/Test (e.g., Cora: 0.05/0.18/0.37) |
| Hardware Specification | Yes | All experiments are conducted on two Nvidia Quadro RTX 6000 GPUs. |
| Software Dependencies | No | The paper mentions using PyTorch and refers to implementations from Du et al. (2019b) and Rong et al. (2019) but does not specify version numbers for PyTorch or any other software libraries used. |
| Experiment Setup | Yes | We conduct experiments on a GCN (Kipf & Welling, 2017), where we apply a width of 1, 000 at each hidden layer and the depth ranging from 2 to 29. Figure 2 shows the training and test accuracy on Cora, Citesser and Pubmed after 300 training epochs. ... For C-Drop Edge, we perform a random hyper-parameter search and fix the edge preserving rate as ρ(G) = |V | 2|E|. |