Figure S21: The neural ensemble extrapolates across rotations/scales

The format and conventions are the same as in Figure S13. A. Scale invariance. The classifier was trained with the neural responses obtained upon presenting objects at the default scale (3 degrees) and testing its performance with the neural responses obtained upon presenting objects at half size or twice the size. The classification performance values reported here show the average between the two scales. B. Rotation invariance. The classifier was trained with the neural responses obtained upon presenting objects at the default rotation (Figure S1) and testing its performance with the neural responses obtained upon presenting objects at 45 and 90 degrees rotation. The classification performance values reported here show the average between the two rotations. Electrodes were selected based on the rv values (ratio of the variance across categories to the variance within categories) using only the training data.