Precedence-Constrained Winter Value for Effective Graph Data Valuation
Authors: Hongliang Chi, Wei Jin, Charu Aggarwal, Yao Ma
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on various datasets and tasks, along with detailed ablation studies and parameter analyses, validate the effectiveness of PC-Winter and provide insights into its behavior. |
| Researcher Affiliation | Collaboration | 1Rensselaer Polytechnic Institute, Troy, NY, United States 2Emory University, Atlanta, GA, United States 3IBM T. J. Watson Research Center, Yorktown Heights, NY, United States |
| Pseudocode | No | The paper describes methods and strategies but does not include any explicitly labeled "Pseudocode" or "Algorithm" blocks or figures with structured steps. |
| Open Source Code | Yes | Code is released at https://github.com/frankhlchi/graph-data-valuation. |
| Open Datasets | Yes | We assess the proposed approach on six real-world benchmark datasets: Cora, Citeseer, and Pubmed Sen et al. (2008), Amazon-Photo, Amazon-Computer, and Coauther-Physics Shchur et al. (2018). |
| Dataset Splits | Yes | Following (Hamilton et al., 2017), we split each graph G into 3 disjoint subgraphs: training graph Gtr, validation graph Gva, and test graph Gte... For the citation networks, we adopt public train/val/test splits in our experiments. For the remaining datasets, we randomly select 20 labeled nodes per class for training, 20% nodes for validation and 20% nodes as the testing set. |
| Hardware Specification | Yes | To address this and to stay within a realistic scope, we cap the computation time at 120 GPU hours on NVIDIA Titan RTX, after which the calculation is terminated. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9). |
| Experiment Setup | Yes | Table 3 includes the hyper-parameters and truncation ratios used for value estimation. Dataset Truncation Ratio Learning Rate Epoch Weight Decay Cora 0.5-0.7 0.01 200 5e-4 ... |