An Aligned Subtree Kernel for Weighted Graphs

Authors: Lu Bai, Luca Rossi, Zhihong Zhang, Edwin Hancock

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on standard datasets demonstrate that our kernel can easily outperform state-of-the-art graph kernels in terms of classification accuracy.
Researcher Affiliation Academia Lu Bai BAILU69@HOTMAIL.COM School of Information, Central University of Finance and Economics, Beijing, China Luca Rossi L.ROSSI@CS.BHAM.AC.UK School of Computer Science, University of Birmingham, Birmingham, UK Zhihong Zhang ZHIHONG@XMU.EDU.CN Software School, Xiamen University, Xiamen, Fujian, China Edwin R. Hancock ERH@CS.YORK.AC.UK Department of Computer Science, University of York, York, UK
Pseudocode Yes The pseudocode of the TI algorithm is shown in Algorithm 1, where the m-sphere neighbourhood of a vertex v D VD is denoted as N(v D) = {u D VD|S(v D, u D) = m} and S(v D, u D) is the shortest path length between v D and u D. Algorithm 1 Vertex labels strengthening procedure
Open Source Code No The paper does not provide any statement or link regarding the public availability of its source code.
Open Datasets Yes BAR31, BSPHERE31 and GEOD31: The SHREC 3D Shape database consists of 15 classes and 20 instances per class, for a total of 300 shapes (Biasotti et al., 2003)... MUTAG: The dataset consists of weighted graphs representing 188 chemical compounds.
Dataset Splits Yes For each kernel, we perform 10-fold cross-validation where the classification accuracy is computed using a C-Support Vector Machine (C-SVM).
Hardware Specification Yes The runtime is measured under Matlab R2011a running on a 2.5GHz Intel 2-Core processor (i.e., i5-3210m).
Software Dependencies Yes The runtime is measured under Matlab R2011a running on a 2.5GHz Intel 2-Core processor (i.e., i5-3210m)... In particular, we make use of the LIBSVM library(Chang & Lin, 2011).
Experiment Setup Yes For our ASK kernel k(M) EM , we set h to 10 and M to 50... For the WLSK kernel and the JTQK kernel, we set the highest dimension (i.e., the highest height of subtrees) of the Weisfeiler-Lehman isomorphism (for the WLSK kernel) and the tree-index method (for the JTQK kernel) to 10. For the DBMK kernel, we set the highest layer of the required DB representation to 10.