View on GitHub

FCTensDec

Implementation of the Higher Order Sparse Mode Reduction (HOSMR), and other dimensionality reduction methods.

Download this project as a .zip file Download this project as a tar.gz file

Associated publication

Multiplex network inference with sparse tensor decomposition for functional connectivity [hal]
Gaëtan Frusque, Julien Jung, Pierre Borgnat, Paulo Gonçalves, IEEE transaction T-SIPN, 03-2020

Desciption

We provide the necessary code to remake the experiments from the article.

You need first to download these toolboxes:

The methods implemented in this library to perform dimensionality reduction are :

1) Non Negative matrix Factorisation (NMF)

2) Non Negative Canonical Polyadic decomposition (nnCP)

3) Singular Value Decomposition (SVD)

4) Higher Order Singular Value Decomposition (HOSVD)

5) Sparse Singular Value Decomposition (sSVD)

6) Higher Order Sparse Mode Reduction (HOSMR)

Overview of the codes provided

FCMain_model_comparison.m: Main code to remake the experiment from the article. This file is divided in five parts. I) Show results from the paper from the dataset Results.mat. II) Build your own experiment using the model from the paper to compare all methods. III) & IV) Perform the full experiment from the paper. V) Experiment from appendix B.

Beside implementations of the dimensionality reduction methods, this main use the following function:

FCMain_seizure_application.m: Shows results using HOSMR method to discover subgraph of FC caracterising the different events of a seizure.

All 6 methods are named respectively FC_NMF.m, FC_nnCP.m, FC_SVD.m, FC_mHOOI.m (for the HOSVD), FC_sSVD.m, FC_HOSMR.m. Their arguments are respectively:

Thank you for your interest.