I. Introduction
Magnetoencephalography (MEG) is completely noninvasive, nonhazardous technology for functional brain mapping. Every current generates a magnetic field, and following the same principle in the nervous system, the longitudinal neuronal current flow generates an associated magnetic field. MEG detects weak extracranial magnetic fields in the brain, and allows determination of their intracranial sources [1], [2], giving a direct information on the brain activity, spontaneously or to a given stimulus. In comparison, computed tomography (CT) or magnetic resonance imaging(MRI) provides structural/anatomical information. Functional magnetic resonance imaging (fMRI), relying on blood flow, oxygenation changes, measures neuronal activity only indirectly [3]. By measuring these magnetic fields, scientists can accurately pinpoint the location of the cells/zones of the brain that produce each field. These spatio-temporal signals are used to study human cognition and, in clinical settings, for preoperative functional brain mapping, epilepsy diagnosis, etc. One common method of collecting functional data involves the presentation of a stimulus to a subject. However, most often, the inherent noise level in the data collection process is large enough to obscure the signal(s) of interest. In order to reduce the level of noise, the stimulus is repeated for as many as 100–500 trials. The trials are temporally aligned based on the timing of the stimulus presentation, and then an average is computed. This ubiquitously used approach works well, but it requires numerous trials. This, in turn, causes subject fatigue and, therefore, limits the number of conditions that can be tested for a given subject.