Home

Extract independent sources from data following an linear / bilinear / linear+bilinear structure such as Event Related Potential (ERPs) with Composite Approximate Joint Diagonalization (CAJD).

Reference: MINING THE BILINEAR STRUCTURE OF DATA WITH APPROXIMATE JOINT DIAGONALIZATION, L. Korczowski, F. Bouchard, C. Jutten, M. Congedo, 2016

INTRODUCTION :

Features (effects):

-estimate the spatial and temporal unmixing matrices of data

Results :

-CAJD can extract independent sources from data following a linear / bilinear / linear+bilinear. Thanks to the composite cost function, it is flexible will still succeed if your data is a combinaison of the models with an unknown pondered contribution. -As CAJD can uses several diversity, it is more robust than BAJD from pure bilinear model. -The EEG evoked sources are extracted even with low resolution EEG (tested on 16 electrodes eeg cap)

Installation

Required

-Matlab 2011b or above for source separation -Matlab 2014b or above for vizualization

Steps

-download the last snapshot
-put in any folder
-add the /lib/ folder to your Matlab path

Tutorial

-run the script_CAJD_tuto.m (tested on windows) for an working exemple on simulated data

Troubleshooting

-please send an email at louis[dot]korczowski[at]gmail[dot]com with the detailed Matlab error.