Next: Experiments
Up: The System
Previous: Blob Alignment in
The model neurons used here use the maximum over all input signals
instead of the sum. The reason is that the sum would mix up many
different signals, while only one can be the correct one, i.e., the
total input would be the result of one correct signal and many
misleading ones. Hence the signal-to-noise ratio would be very
low. We have observed an example where even a model identical to the
image was not picked up as the correct one, because the sum over all
the accidental input signals favored a completely different-looking
person. For that reason we introduced the maximum input function,
which is reasonable since the correct signal is likely to be the
strongest one. The maximum rule has the additional advantage that
the dynamic range of the input into a single cell does not vary much
when the connectivity develops, whereas the signal sum would decrease
significantly during synaptic re-organization and let the blobs loose
their alignment.
Next: Experiments
Up: The System
Previous: Blob Alignment in