LifangHe / SDM14_DuSK

DuSK: A Dual Structure-preserving Kernel for Supervised Tensor Learning with Applications to Neuroimages

This is sample implementation of DuSK: A Dual Structure-preserving Kernel for Supervised Tensor Learning with Applications to Neuroimages. Please cite as

Lifang He, Xiangnan Kong, Philip S. Yu, Ann B. Ragin, Zhifeng Hao, Xiaowei Yang
"DuSK: A Dual Structure-preserving Kernel for Supervised Tensor Learning with Applications to Neuroimages"
Proceedings of the 14th SIAM International Conference on Data Mining (SDM14), 2014.
http://.pdf

[Dependencies]

CP tensor factorization Toolbox :

This function needs the CP tensor factorization toolbox (default is cp3_alsls)

SVM solver:

Libsvm toolbox (default is libsvm-3.17)

[Table of Contents]

Main.m : demonstrate the algorithms on a CP factorization dataset and show the usages

Divide.m : Divide the data into k-fold

TrainAverAcc.m : train optimal support higher-order tensor machine with DuSK kernel

Ker_DuSK.m : calculate the DuSK kernel (RBF or linear).

[Dataset]

Data_ADNI.mat : the CP factorization result of ANDI dataset

You can factorize each dataset by cp3_alsls toolbox, like Data_ADNI

[Question?]

Please send your questions to [e-mail]