Figure S3.
Illustration
of how visual based stimulus information can lead to categorization
decoding accuracy even when there is no abstract category information
in the population of neurons. A, an illustration of 4
hypothetical neurons’ responses to two images of dogs and two images of
cats. Each neuron fires action potentials at a high rate to
just
one of images; thus each neuron can be thought of as being visually
selective but not selective to the abstract categories. B, if
training is done using trials from all when all 4 cat and dog images
are shown, then one can obtain perfect cat/dog classification accuracy,
even though these hypothetical neurons are only selective to visual
features of the stimuli (and even though neural responses are
noisy). C, if the training is done using responses from just
one
cat and one dog image, and the testing is done using responses to the
other cat and dog images, then if the neurons are only respond to
visual properties of the stimuli, classification performance will be at
chance.
Home