Natural Graph Networks

Authors: Pim de Haan, Taco S. Cohen, Max Welling

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 7 Experiments, Icosahedral MNIST In order to experimentally show that our method is equivariant to global symmetries, and increases expressiveness over an invariant message passing network (GCN), we classify MNIST on projected to the icosahedron, as is done in Cohen et al. [2019]. We evaluate our model with GCN2 message parametrization on a standard set of 8 graph classification benchmarks.
Researcher Affiliation Collaboration Pim de Haan Qualcomm AI Research University of Amsterdam QUVA Lab Taco Cohen Qualcomm AI Research Max Welling Qualcomm AI Research University of Amsterdam
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code, a specific repository link, or an explicit code release statement for the methodology described.
Open Datasets Yes Icosahedral MNIST In order to experimentally show that our method is equivariant to global symmetries, and increases expressiveness over an invariant message passing network (GCN), we classify MNIST on projected to the icosahedron, as is done in Cohen et al. [2019]. We evaluate our model with GCN2 message parametrization on a standard set of 8 graph classification benchmarks from Yanardag and Vishwanathan [2015].
Dataset Splits Yes We use the 10-fold cross validation method as described by Zhang et al. [2018] and report the best averaged accuracy across the 10-folds, as described by Xu et al. [2018].
Hardware Specification No The paper states, 'These experiments were run on QUVA machines,' which is too vague to determine specific hardware details like GPU/CPU models or memory amounts.
Software Dependencies No The paper mentions names of methods like 'GCN2' but does not specify software packages or libraries with version numbers required for reproducibility.
Experiment Setup Yes In our experiments, we choose Gp to contain all nodes in G that are at most k edges removed from p, for some natural number k, and all edges between these nodes. Similarly, we pick for edge pp, qq P Ep Gq neighbourhood Gpq containing all nodes at most k edges removed from p or q and all edges between these nodes. In all experiments, we chose k 1, unless otherwise noted.