G$^2$N$^2$ : Weisfeiler and Lehman go grammatical

Authors: Jason Piquenot, Aldo Moscatelli, Maxime Berar, Pierre Héroux, Romain Raveaux, Jean-Yves RAMEL, Sébastien Adam

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

Reproducibility Variable Result LLM Response
Research Type Experimental Through various experiments, we demonstrate the superior efficiency of G2N2 compared to other 3-WL GNNs across numerous downstream tasks. This section is dedicated to the experimental validation of both the framework and G2N2.
Researcher Affiliation Academia LITIS Lab, University of Rouen Normandy, France LIFAT Lab, University of Tours, France
Pseudocode No The paper describes procedures using figures and text but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or link indicating the release of open-source code for the described methodology.
Open Datasets Yes We use a graph regression benchmark called QM9 which is composed of 130K small molecules (Ramakrishnan et al. (2014); Wu et al. (2018)). For graph classification, we evaluate G2N2 on the classical TUD benchmark (Morris et al. (2020)).
Dataset Splits Yes As in Maron et al. (2019a), the dataset is randomly split into training, validation, and test sets with a respective ratio of 0.8, 0.1 and 0.1.
Hardware Specification No The paper states that "The mean epoch duration is measured on the same device" but does not provide any specific details about the hardware used for experiments.
Software Dependencies No The paper does not specify the version numbers for any software dependencies used in the experiments.
Experiment Setup Yes In the experiments, all the linear blocks of a layer are set at the same width S(l) = b(l) = b(l) = b(l) diag. This means that MLP(l) M takes as input a third order tensor of dimensions n n 4S(l) and MLP(l) Vc takes as input a matrix of dimensions n 2S(l). At each layer, the MLP depth is always 2 and the intermediate layer doubled the input dimension. The parameter setting for each of the 6 experiments related to these datasets can be found in Table 5 of appendix C.