Figure 7.

Comparison between different linear discrimination methods for subject JMB. (a) Using sensorwise difference of mean signals for two conditions as weights. (b) Same as (a) but the weight of each sensor is normalized by the variance of the signal at that sensor. (c) Regularized Fisher linear discriminant analysis. The plots depict error estimate of the classifier as a function of time slice of MEG signal to which it was applied. Dotted lines denote 1-std-wide margin around the estimate. The maps depict distribution of weights over the scalp (flattened helmet viewed from above) at the time slice that yields best separation (marked by blue arrow).