Log in

Login to your account

Username *
Password *
Remember Me
From Brain Science to Intelligent Machines

Dealing with non-stationary EEG signals in brain-computer interface

Date: Friday 1/3/2013
Venue: MS020
Time: 12.00 pm
Speaker: Ms. Mahnaz Arvaneh
Affiliation: Institute for Infocomm Research (I2R), Singapore

Dealing with non-stationary EEG signals in brain-computer interface

Ms. Mahnaz Arvaneh

Institute for Infocomm Research (I2R), Singapore



A major challenge in EEG-based brain-computer interfaces (BCIs) is the inherent non-stationarities in the EEG data. Variations of the signal properties from intra and inter sessions often lead to deteriorated BCI performances. To address this issue, we proposed two novel algorithms.
The first algorithm focuses on intra-session non-stationarities by extracting features that are robust and invariant. This was achieved by a novel spatial filtering algorithm, called KLCSP. The proposed KLCSP algorithm simultaneously maximizes the discrimination between the class means, and minimizes the within-class dissimilarities measured by a loss function based on the Kullback-Leibler (KL) divergence. The proposed KLCSP algorithm is completely data-driven, and only uses the train data to make the features robust and invariant against non-stationarities and variations.
The second algorithm focuses on inter-session non-stationarities by proposing a novel data space adaptation technique, called EEG data space adaptation (EEG-DSA). The EEG-DSA algorithm linearly transforms the EEG data from the evaluation session, such that the distribution difference to the training session is minimized. Using the KL divergence criterion, two versions of the EEG-DSA algorithm are proposed: the supervised version when labelled data are available in the evaluation session, and the unsupervised version when labelled data are not available. By adapting the EEG data space directly, the proposed EEG-DSA algorithm is not limited to any specific BCI models. In contrast, it can be applied with other techniques which adapt the feature or classifier spaces. In addition, the proposed algorithm can be potentially applied on other spaces, such as ECoG, MEG etc.

Short biography:

Mahnaz Arvaneh received the B.Sc. degree in electrical engineering from K. N. Toosi University of Technology, Tehran, Iran, and the M.Sc. degree in control engineering from Ferdowsi University of Mashhad, Iran, in 2005 and 2007, respectively. She is currently pursuing the Ph.D. degree with Singapore Nanyang Technological University. Since April 2009, she is an attached student in Institute for Infocomm research, Agency for Science, Technology and Research, Singapore. In 2012, she was a visiting student with the Electrical Engineering Department, National University of Ireland, Maynooth, Ireland. Her current research interests include brain–computer interfaces, signal processing, machine learning, and pattern recognition.