Identity decoding for images within the same category vs.
mixed categories
The
figures below show the results from decoding the identity of images
using images derived from three prototypes at a time (as proposed by
one of
the anonymous
reviewers of the paper).
A comparison is made between the average of this decoding for
the
within category prototypes ([C1, C2, C3] and [D1, D2, D3]) (green line)
to the average of the 18 other mixed category permutations ([C1 C2,
D1], [C1 C2 D2], ..., [C3 D2 D3]) (magenta line). Since the
readout is for the identity of the 21 images, chance is 1/21.
As
can be seen in the figures, for both PFC and ITC there is a significant
amount of visual information present, and the potential presence of
'abstract' category information in the within-category condition does
not seem to significantly degrade the ability of the classifier to
decode the 'visually based' identity information.
Home