shouhengtuo / mkmtl

Multikernel Multitask Learning code. We refer to the paper for details about the model and the optimization algorithms: Xiaoli Liu, Peng Cao, Jinzhu Yang, Dazhe Zhao. Linearized and Kernelized Sparse Multitask Learning for Predicting Cognitive Outcomes in Alzheimer’s Disease[J]. Computational and mathematical methods in medicine, 2018. DOI=10.1155/2018/7429782

Multikernel Multitask Learning

This repository contains two MATLAB implementations of the MKMTL algorithm proposed in the paper Linearized and Kernelized Sparse Multitask Learning for Predicting Cognitive Outcomes in Alzheimer’s Disease.

Overview

Multikernel Multitask Learning can model the cognitive scores as nonlinear functions of neuroimaging measures. Two nonlinear multikernel-based multitask learning methods was proposed in this work to exploit and investigate the nonlinear relationship between MRI measures and cognitive scores. lql1-MKMTL learns a common kernel representation by imposing sparsity constraint on the kernel weight. It assumes that few base kernels are important for the tasks and encourages a linear combination of only few kernels and assumes few selected kernels are similar across the tasks. l2,1lq-MKMTL builds the nonlinear relationship for the features and task to capture the interrelation ships between different cognitive measures over the feature space.

This code has been tested only in both Linux and Mac.

How to run?

We created the file exam_lql1MKMTL.m and exam_l21lqMKMTL.m to show how to run lql1-MKMTL and l21lq-MKMTL codes, respectively.

Structure of the input data files

In order to run the code the input data files containing the training and test data must follow a specific format. The lql1MKMTL() function, which is the core algorithm of lql1-MKMTL, receives four matrices, X (covariate matrix) n x p with the number of n samples and p covariates, and Y (response matrix) n x k with k tasks; and kernel information cells. For algorithm l21lq-MKMTL, the parameters() function is firstly used to calculate the parameters. Then the trainl21lq() function, which is the core algorithm, employed to train the model.

How to cite it?

If you like it and want to cite it in your papers, you can use the following:

#!latex

@article{liu2018linearized,
  title={Linearized and Kernelized Sparse Multitask Learning for Predicting Cognitive Outcomes in Alzheimer’s Disease},
  author={Liu, Xiaoli and Cao, Peng and Yang, Jinzhu and Zhao, Dazhe},
  journal={Computational and mathematical methods in medicine},
  volume={2018},
  year={2018},
  publisher={Hindawi}
}

Have a question?

If you found any bug or have a question, don't hesitate to contact me:

[Xiaoli Liu] email: neuxiaoliliu -at- gmail -dot- com