(
A) Maps of the spatial attributes of neural firing for each of 42 simultaneously recorded neurons in the ACC. Red indicates highest likelihood of unit firing. (
B) Prediction error versus the width of the smoothing kernel for all seven sessions analyzed. The dashed line indicates the width used for decoding analyses in
Figure 2. (
C) Histogram of the prediction errors in Cartesian coordinates for one representative session. (
D) Prediction error (RMSE) as a function of the number of qualified units recorded for several sessions from each animal (symbols and colors). The prediction error increased with fewer units. To test if the poor decoding performance for some subjects was due to the low number of recorded units, we tested if downsampling the number units from another animal with 40 units would result in similar poor performance. The prediction error increased super-linearly as the downsampling was increased (grey ‘+’) and is in-line with the poor predictions from the two animals with few units. (
E) The prediction error achieved by using the most informative units, as determined by a greedy selection process for the two animals with the most units. Colors correspond to those in panel D.