Fine-Tuning Graph Neural Networks via Graph Topology Induced Optimal Transport

Authors: Jiying Zhang, Xi Xiao, Long-Kai Huang, Yu Rong, Yatao Bian

IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate GTOT-Tuning on eight downstream tasks with various GNN backbones1 and demonstrate that it achieves state-of-the-art fine-tuning performance for GNNs. The experiments conducted on eight different datasets with various GNN backbones show that GTOT-Tuning achieves the best performance among all baselines, validating the effectiveness and generalization ability of our method.
Researcher Affiliation Collaboration Jiying Zhang1,2 , Xi Xiao2 , Long-Kai Huang1 , Yu Rong1 and Yatao Bian1 1Tencent AI Lab 2Shenzhen International Graduate School, Tsinghua University
Pseudocode Yes Algorithm 1 Computing Masked Wasserstein Distance
Open Source Code Yes Code: https://github.com/youjibiying/GTOT-Tuning.
Open Datasets Yes In addition, eight binary classification datasets in Molecule Net [Wu et al., 2018] serve as downstream tasks for evaluating the finetuning strategies, where the scaffold split scheme is used for dataset split.
Dataset Splits No The paper mentions that 'the scaffold split scheme is used for dataset split' for MoleculeNet, but it does not provide specific quantitative details like percentages or sample counts for training, validation, or test sets in the main body. It refers to an appendix for more details, which is not available.
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., GPU/CPU models, memory details) used to run its experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., 'Python 3.8', 'PyTorch 1.9', 'CUDA 11.1') needed to replicate the experiment. It refers to models and datasets but not the specific software environment for running them.
Experiment Setup No The paper mentions that 'λ is a hyper-parameter for balancing the regularization with the main loss function' and details the loss function. However, it does not provide specific numerical values for hyperparameters such as λ, learning rates, batch sizes, or optimizer settings in the main text. It defers some details to an appendix, which is not available.