Graph Out-of-Distribution Detection Goes Neighborhood Shaping

Authors: Tianyi Bao, Qitian Wu, Zetian Jiang, Yiting Chen, Jiawei Sun, Junchi Yan

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental results show the competitiveness of the proposed model across multiple datasets, as evidenced by up to a 15% increase in the AUROC and a 50% decrease in the FPR compared to existing state-of-the-art methods.
Researcher Affiliation Academia 1School of Artificial Intelligence & Department of Computer Science and Engineering & Mo E Lab of AI, Shanghai Jiao Tong University, Shanghai, China.
Pseudocode No The paper does not contain explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement about releasing its source code or a link to a code repository for the methodology described.
Open Datasets Yes We ground our experiments on six prominent real-world datasets, frequently employed in node classification benchmarks: Twitch-Explicit (Rozemberczki & Sarkar, 2021), ogbn-Arxiv (Hu et al., 2020), Amazon-Photo (Mc Auley et al., 2015), Coauthor-CS (Sinha et al., 2015), Coauthor Physics (Shchur et al., 2018), and Cora (Sen et al., 2008).
Dataset Splits Yes For ID data, we employed the conventional random splits method (1:1:8 for training/validation/testing) as suggested by Kipf & Welling.
Hardware Specification Yes Most of the experiments run with an NVIDIA 2080Ti with 11GB memory, except for cases where the model requires larger GPU memory, for which we use an NVIDIA 3090 with 24GB memory for experiments.
Software Dependencies Yes Our implementation is based on Ubuntu 16.04, Cuda 11.0, Pytorch 1.13.0, and Pytorch Geometric 2.3.1.
Experiment Setup Yes We set the number of propagation steps k according to the size of the graph datasets, namely, k equal to 5 or 10 for most settings, and the propagation coefficient α = 0.5. For fair comparison, the GCN model with layer depth 2 and hidden size 64 is used as the backbone encoder for all the OOD discriminators. We detail the default hyper-parameters utilized across all scenarios, as delineated in Tab. 5.