Graph Kernels: A Survey

Authors: Giannis Nikolentzos, Giannis Siglidis, Michalis Vazirgiannis

JAIR 2021 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Furthermore, we perform an experimental evaluation of several of those kernels on publicly available datasets, and provide a comparative study. ... In Section 7, we experimentally evaluate the performance of many graph kernels on several widely-used graph classification benchmark datasets.
Researcher Affiliation Academia Giannis Nikolentzos EMAIL LIX, Ecole Polytechnique Palaiseau, 91120, France; Ioannis Siglidis EMAIL LIGM, Ecole des Ponts, Universit e Gustave Eiffel, CNRS Marne-la-Vall ee, 77420, France; Michalis Vazirgiannis EMAIL LIX, Ecole Polytechnique Palaiseau, 91120, France. All listed institutions are academic or public research organizations in France, and the email domains also indicate academic affiliations.
Pseudocode No The paper describes algorithms and procedures (e.g., Geometric Random Walk Kernel, Weisfeiler-Lehman Framework, Neighborhood Hash Kernel) in descriptive text, but it does not include formally structured pseudocode blocks or algorithms with numbered steps or code-like formatting.
Open Source Code No Specifically, we made use of the Gra Ke L library which contains implementations of a large number of graph kernels (Siglidis et al., 2020). The authors state that they *used* an existing open-source library (GraKeL) for their experimental comparison, rather than releasing their own source code specifically for the methodologies described within this survey paper.
Open Datasets Yes All datasets are publicly available (Kersting et al., 2016). ... Kersting, K., Kriege, N. M., Morris, C., Mutzel, P., & Neumann, M. (2016). Benchmark data sets for graph kernels.. http://graphkernels.cs.tu-dortmund.de.
Dataset Splits Yes Therefore, we perform 10-fold cross-validation to obtain an estimate of the generalization performance of each method. For the common datasets, we use the splits (and results) provided by Errica et al. (2020).
Hardware Specification Yes All experiments were performed on a cluster of 80 Intel Xeon CPU E7 4860 @ 2.27GHz with 1TB RAM.
Software Dependencies No Specifically, we made use of the Gra Ke L library which contains implementations of a large number of graph kernels (Siglidis et al., 2020). We also employed a Support Vector Machine (SVM) classifier and in particular, the LIB-SVM implementation (Chang & Lin, 2011). While specific libraries (GraKeL, LIB-SVM) are mentioned, their version numbers are not explicitly provided.
Experiment Setup Yes Within each fold, the parameter C of the SVM and the hyperparameters of the kernels (see below) and GNNs were chosen based on a validation experiment on a single 90% 10% split of the training data. We chose the value of parameter C from {10 7, 10 5, . . . , 105, 107}. Moreover, we normalized all kernel values as follows ˆk(Gi, Gj) = k(Gi,Gj)/√k(Gi,Gi) k(Gj,Gj) for any graphs Gi, Gj. ... The values of the different hyperparameters of the kernels are shown in Table 4.