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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
HubGT: Fast Graph Transformer with Decoupled Hierarchy Labeling
Authors: Ningyi Liao, Zihao Yu, Siqiang Luo, Gao Cong
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that Hub GT offers efficient computation and mini-batch capability over existing GT designs on large-scale datasets while achieving top-tier effectiveness. Our code is available at: https://github.com/gdmnl/Hub GT. |
| Researcher Affiliation | Academia | Ningyi Liao Nanyang Technological University EMAIL Zihao Yu Nanyang Technological University EMAIL Siqiang Luo Nanyang Technological University EMAIL Gao Cong Nanyang Technological University EMAIL |
| Pseudocode | Yes | Our Algorithms 1 to 3 are detailed in Appendix A.2 |
| Open Source Code | Yes | Our code is available at: https://github.com/gdmnl/Hub GT. |
| Open Datasets | Yes | We focus on the node classification task on 14 benchmark datasets covering both homophily [37, 38, 39] and heterophily [40, 41]. Compared to conventional graph learning tasks used in GT studies, this task requires learning on large single graphs, which is suitable for assessing model scalability. We follow common data processing and evaluation protocols as detailed in Appendix C. |
| Dataset Splits | Yes | We employ 60/20/20 random data splitting percentages for training, validation, and testing sets, respectively, except for OGBN-MAG, where the original split is used. |
| Hardware Specification | Yes | Evaluation is conducted on a server with 32 Intel Xeon CPUs (2.4GHz), an Nvidia A30 GPU (24GB memory), and 512GB RAM. |
| Software Dependencies | No | The paper does not explicitly state software dependencies with specific version numbers in the provided text. |
| Experiment Setup | Yes | For network architectural hyperparameters, we use L = 4 Transformer layers with NH = 8 heads and F = 128 hidden dimension for our Hub GT model across all experiments. The dropout rates for inputs (features and bias) and intermediate representation are 0.1 and 0.5, respectively. The Adam W optimizer is used with a learning rate of 10^-4. The model is trained with 300 epochs with early stopping. |