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..
Parameter-Free Hypergraph Neural Network for Few-Shot Node Classification
Authors: Chaewoon Bae, Doyun Choi, Jaehyun Lee, Jaemin Yoo
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct comprehensive experiments on 11 real-world hypergraphs to verify the effectiveness of ZEN. |
| Researcher Affiliation | Academia | School of Electrical Engineering Korea Advanced Institute of Science and Technology (KAIST) EMAIL |
| Pseudocode | Yes | Algorithm 1 Approximation via random walks |
| Open Source Code | Yes | Our code and datasets are fully available at https://github.com/chaewoonbae/ZEN. |
| Open Datasets | Yes | Our code and datasets are fully available at https://github.com/chaewoonbae/ZEN. We evaluate ZEN on a total of 11 real-world hypergraph graphs. To assess predictive performance and computational efficiency, we use 10 standard benchmarks: Cora, Citeseer, Pubmed, Cora_CA, 20News100, Model Net40, Congress, Walmart, Senate, and House, following prior work [15, 23]. For interpretability analysis, we use Zoo [15], a small dataset whose feature attributes are semantically interpretable. Detailed dataset statistics are provided in Table 2. |
| Dataset Splits | Yes | For each split, we allocate 5 labeled nodes per class for training, and another 5 nodes per class for validation [12, 27], making 5-shot node classification. The remaining nodes are used for testing. |
| Hardware Specification | Yes | All our experiments are conducted with NVIDIA RTX A6000 and AMD EPYC 9354. |
| Software Dependencies | No | The paper mentions 'Adam optimizer' but does not specify version numbers for any key software components or libraries used in the implementation. |
| Experiment Setup | Yes | All baselines are trained using the Adam optimizer with no weight decay, and we conduct a grid search over 72 hyperparameter configurations: lr {10 2, 10 3, 10 4}, epochs {50, 100, 150, 200}, num_layers {1, 2}, hidden_dim {64, 128, 256}. In contrast, ZEN requires no training hyperparameters. Instead, we search over 55 combinations of propagation coefficients (α0, α1, α2) uniformly sampled from the 2-simplex, yielding a comparable hyperparameter space size to that of baselines. |