Subgroup Generalization and Fairness of Graph Neural Networks
Authors: Jiaqi Ma, Junwei Deng, Qiaozhu Mei
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Despite enormous successful applications of graph neural networks (GNNs), theoretical understanding of their generalization ability, especially for node-level tasks where data are not independent and identically-distributed (IID), has been sparse. The theoretical investigation of the generalization performance is beneficial for understanding fundamental issues (such as fairness) of GNN models and designing better learning methods. In this paper, we present a novel PAC-Bayesian analysis for GNNs under a non-IID semi-supervised learning setup. Moreover, we analyze the generalization performances on different subgroups of unlabeled nodes, which allows us to further study an accuracy-(dis)parity-style (un)fairness of GNNs from a theoretical perspective. Under reasonable assumptions, we demonstrate that the distance between a test subgroup and the training set can be a key factor affecting the GNN performance on that subgroup, which calls special attention to the training node selection for fair learning. Experiments across multiple GNN models and datasets support our theoretical results4. |
| Researcher Affiliation | Academia | Jiaqi Ma jiaqima@umich.edu Junwei Deng junweid@umich.edu Qiaozhu Mei qmei@umich.edu School of Information, University of Michigan, Ann Arbor, Michigan, USA Equal contribution. Department of EECS, University of Michigan, Ann Arbor, Michigan, USA |
| Pseudocode | No | No section or figure explicitly labeled 'Pseudocode' or 'Algorithm' was found, nor were there structured steps formatted as code or an algorithm. |
| Open Source Code | Yes | Code available at https://github.com/TheaperDeng/GNN-Generalization-Fairness. |
| Open Datasets | Yes | The main paper reports the results on a small set of datasets (Cora, Citeseer, and Pubmed [37, 47]). Results on more datasets, including large-scale datasets from Open Graph Benchmark [17], are shown in Appendix C. |
| Dataset Splits | Yes | Following common GNN experiment setup [38], we randomly select 20 nodes in each class for training, 500 nodes for validation, and 1,000 nodes for testing. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for the experiments, such as GPU or CPU models. |
| Software Dependencies | No | The paper mentions using "implementations by Deep Graph Library [43]" but does not specify a version number for this library or any other software dependencies with their versions. |
| Experiment Setup | No | While the paper describes the general experimental setup, such as the number of nodes for training, validation, and testing, and the GNN models used, it does not provide specific hyperparameter values (e.g., learning rate, batch size, number of epochs) or detailed system-level training configurations in the main text. |