Kernelized Support Tensor Machines

Authors: Lifang He, Chun-Ta Lu, Guixiang Ma, Shen Wang, Linlin Shen, Philip S. Yu, Ann B. Ragin

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experimental results on realworld neuroimaging datasets show the superiority of KSTM over the state-of-the-art techniques.
Researcher Affiliation Academia 1Department of Computer Science, University of Illinois at Chicago, Chicago, IL, USA 2Institute for Computer Vision, Shenzhen University, Shenzhen, China 3Institute for Data Science, Tsinghua University, Beijing, China 4Department of Radiology, Northwestern University, Chicago, IL, USA.
Pseudocode Yes Algorithm 1 Learning KSTMs
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes Alzheimer’s Disease (ADNI): This dataset is collected from the Alzheimer’s Disease Neuroimaging Initiative1. (http://adni.loni.usc.edu/); Human Immunodeficiency Virus Infection (HIV): This dataset is collected from Chicago Early HIV Infection Study in Northwestern University (Wang et al., 2011); Attention Deficit Hyperactivity Disorder (ADHD): This dataset is collected from ADHD-200 global competition dataset4. (http://neurobureau.projects.nitrc.org/ADHD200/)
Dataset Splits Yes We perform 5-fold cross-validation and use classification accuracy as the evaluation measure.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments, such as GPU/CPU models or memory specifications.
Software Dependencies Yes We applied SPM8 2 and REST3 to preprocess the data. (Footnote 2: http://www.l.ion.ucl.ac.uk/spm/software/spm8/; Footnote 3: http://resting-fmri.sourceforge.net); For the evaluation where SVM is needed, we apply Lib SVM (Chang & Lin, 2011), a widely used implementation of SVM, with RBF kernel as the classifier.
Experiment Setup Yes The optimal parameters for all methods are determined by grid search. The optimal trade-off parameter is selected from C {2 8, 2 7, , 28}, the kernel width parameter is selected from σ {2 8, 2 7, , 28}, the optimal rank R is selected from {1, 2, , 10}, and the regularized factorization parameter is selected from γ {2 8, 2 7, , 28}.