Classication of semantic memories using multitaper spectral estimation

Abstract: The research on classication of semantic memories is still very young. Several methods have been tested ranging from magnetic resonance imaging (MRI) to electrocorticog- raphy (ECoG). This report describes an alternative way of classifying signals collected from an electroencephalogram (EEG) into categories using the Thomson multitaper method of spectral estimation, as well as a logistic regression model. The aim for this report is to expand the research eld with an approach that complements the current options of classication. Data was distributed from the department of Psychology at Lund University, and the experimental paradigm was to classify three types of semantic memories (faces, landmarks and objects) based on their neural patterns. Based on the cross-validation from the mentioned methods, a classier could successfully be trained for the "faces" and "landmarks" categories with an average success rate of 55% and 51% respectively. The classier accurately responded to the onset of the stimuli (p < 0:001 for faces, p = 0:015 for landmarks). No classier for the "objects" category could be trained using this method. These results indicate that the multitaper method of spec- tral estimation can be useful in detecting neural patterns. Several ways to rene these methods are discussed.