Abstract—The accessibility to Electroencephalogram (EEG) recording systems has enabled the healthcare providers to record the brain activity of patients under treatment, during multiple sessions. Thus brain changes can be observed and evaluated. It has been shown in many studies that the EEG data are never exactly the same when recordings are done in different sessions inducing a shift between the data of multiple sessions. This shift is induced due to the changes in parameters such as: the physical /mental state of the patient, the ambient environment, location of the electrodes, and impedance of the electrodes. The shift can be modelled as a covariate shift between multiple sessions. However, the algorithms that have been developed to tackle this shift assume the presence of training as well as testing data apriori to calculate the importance weights which are then used in the learning algorithm to reduce the mismatch. This major problem makes them impractical. In this paper, we tackle this, using marginalized stacked denoising autoencoder (mSDAs) while using the data from seven healthy subjects recorded over eightsessions distributed over four weeks. We compare our results with kernel mean matching, a popular approach for covariate shift adaption. Using support vector machines for classification and reduced complexity of mSDA, we get promising accuracy.

Index Terms—Electroencephalogram, transfer learning, marginalized stacked denoising autoencoders, covariate shift adaptation.

Farah Abid1, Ali Hassan is with College of Electrical and Mechanical Engineering National University of Sciences and Technology, Pakistan (email: alihassan@ceme.nust.edu.pk).
Anum Abid is with University of Engineering and Technology Taxilla, Pakistan.
Imran Khan Niazi is with Centre for Chiropractic Research, New Zealand College of Chiropractic, Auckland 1060, New Zealand.
Mads Jochumsen is with Center for Sensory-Motor Interaction, Department of Health Science and Technology, Aalborg University, Denmark.

Cite: Farah Abid, Ali Hassan, Anum Abid, Imran Khan Niazi, Mads Jochumsen, "Transfer Learning for Electroencephalogram Signals," International Journal of Bioscience, Biochemistry and Bioinformatics vol. 7, no. 3, pp. 143-152, 2017.