Graph Homomorphism Convolution

Authors: Hoang Nguyen, Takanori Maehara

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

Reproducibility Variable Result LLM Response
Research Type Experimental 4. Experimental results
Researcher Affiliation Academia 1RIKEN Center for Advanced Intelligence Project, Tokyo, Japan 2Tokyo Institute of Technology, Tokyo, Japan.
Pseudocode Yes Algorithm 1 Compute hom(F, (G, x)) Input: target graph G, pattern graph F, vertex features x function recursion(current, visited) hom x x for y in F.neighbors(current) do if y = visited then hom y recursion(y, current) aux [P hom y[G.neighbors(i)] for i in V (G)] hom x hom x aux (element-wise mult.) end if end for return hom x end function Output: P recursion(0, -1)
Open Source Code Yes More details for hyperparamters tuning and source code is available in the Supplementary Materials.
Open Datasets Yes We select 3 datasets from the TU Dortmund data collection (Kersting et al., 2016): MUTAG dataset (Debnath et al., 1991), IMDB-BINARY, and IMDB-MULTI (Yanardag & Vishwanathan, 2015).
Dataset Splits Yes We report the 10-folds cross-validation accuracy scores and standard deviations in Table 2.
Hardware Specification No The paper reports runtime results in Figure 2 but does not provide any specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments.
Software Dependencies No The paper does not explicitly list software dependencies with version numbers in the main text. It mentions that 'source code is available in the Supplementary Materials', which might contain this information, but it is not directly provided in the paper.
Experiment Setup Yes Except for setting the number of epochs for GIN to be 50, we use the default hyperparameters provided by the original papers.