Learning Invariant Graph Representations for Out-of-Distribution Generalization

Authors: Haoyang Li, Ziwei Zhang, Xin Wang, Wenwu Zhu

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on both synthetic and real-world datasets demonstrate the superiority of our method against state-of-the-art baselines under distribution shifts for the graph classification task.
Researcher Affiliation Academia Haoyang Li, Ziwei Zhang, Xin Wang , Wenwu Zhu Tsinghua University lihy18@mails.tsinghua.edu.cn, {zwzhang,xin_wang,wwzhu}@tsinghua.edu.cn
Pseudocode Yes Training Procedure. We present the pseudocode of GIL in Appendix.
Open Source Code Yes 3. If you ran experiments... (a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes]
Open Datasets Yes Datasets. We adopt one synthetic dataset with controllable ground-truth environments and four real-world benchmark datasets for the graph classification task. SP-Motif: Following [13, 16]... MNIST-75sp [17]... Graph-SST2 [19]... Open Graph Benchmark (OGB) [20]: We consider two datasets, MOLSIDER and MOLHIV.
Dataset Splits Yes The hyperparameter λ in Eq. (8) is chosen from {10 5, 10 3, 10 1}. The number of clusters in Eq. (5) is chosen from [2, 4]. They are tuned on the validation set. We report the mean results and standard deviations of five runs.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, memory, or specific cloud computing instances used for experiments.
Software Dependencies No The paper mentions using GNNs and READOUT functions and states, "The adopted GNNs and READOUT functions including GNNM, GNNV, GNNI, READOUTV, and READOUTI are listed in Appendix." It does not specify specific software versions (e.g., PyTorch 1.9, Python 3.8, CUDA 11.1) that would be needed for replication.
Experiment Setup Yes The hyperparameter λ in Eq. (8) is chosen from {10 5, 10 3, 10 1}. The number of clusters in Eq. (5) is chosen from [2, 4]. They are tuned on the validation set.