Figure 4.

Offline decoding performance. (A) Single DOF reconstructions representative of average error on separate training runs for each DOF and subject. Correlations across the entirety of each training run using 5-fold cross validation are shown. (B) Cross validated classification of finger group during onset of flex movements using a support-vector machine. Threshold crossings were binned in a sliding window that was updated every 20 ms. The feature vector was formed from multiple time bins for all available channels.