Revisiting Robustness in Graph Machine Learning

Authors: Lukas Gosch, Daniel Sturm, Simon Geisler, Stephan Günnemann

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Using Contextual Stochastic Block Models (CSBMs) and real-world graphs, our results uncover: i) for a majority of nodes the prevalent perturbation models include a large fraction of perturbed graphs violating the unchanged semantics assumption; ii) surprisingly, all assessed GNNs show over-robustness that is robustness beyond the point of semantic change.
Researcher Affiliation Academia Lukas Gosch, Daniel Sturm, Simon Geisler, Stephan G unnemann Department of Computer Science & Munich Data Science Institute Technical University of Munich {l.gosch, da.sturm, s.geisler, s.guennemann}@tum.de
Pseudocode No Not found. The paper describes methods in text but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes The source code, together will all experiment configuration files of all our experiments can be found on the project page: https://www.cs.cit.tum.de/daml/revisiting-robustness/
Open Datasets Yes Using Contextual Stochastic Block Models (CSBMs) and real-world graphs... CORA (Sen et al., 2008)... Cora-ML (Bojchevski & G unnemann, 2018), Citeseer (Sen et al., 2008), Pubmed (Sen et al., 2008) and ogbn-arxiv (Hu et al., 2020) are selected.
Dataset Splits Yes We use an 80%/20% train/validation split on the nodes. ... for all datasets except ogbn-arxiv, 40 nodes per class are randomly selected as validation and test nodes.
Hardware Specification No Not found. The paper does not specify any hardware details like GPU/CPU models, memory, or specific computing environments used for the experiments.
Software Dependencies No Not found. The paper mentions software like Adam optimizer but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes We train all model for 3000 epochs with a patients of 300 epochs using Adam (Kingma & Ba, 2015) and explore learning rates [0.1, 0.01, 0.001] and weight decay [0.01, 0.001, 0.001] and additionally for MLP: We use a 1 (Hidden)-Layer MLP and test hidden dimensions [32, 64, 128, 256] and dropout [0.0, 0.3, 0.5].