orientation selectivity greater than 1

[kalok87]

Dear developers,

While trying to replicate your simulation of GCAL using the notebook (gcal.ipynb) you provided, I observed than some units in the V1 have selectivity greater than 1 (the biggest selectivity is even equal to 2) which is somehow counter-intuitive.

The selectivity of V1 is simply obtained by using:

RunProgress().run(1000) save_plotgroup('Activity') save_plotgroup('Orientation Preference') p = topo.sim.V1.views.maps.OrientationPreference.last.data s = topo.sim.V1.views.maps.OrientationSelectivity.last.data

After about 50000 iterations, I use np.max(s) to see the maximal selectivity and than I found some values are bigger than 1.

According to the paper, the selectivity cannot be greater than one, right? So what is the reason for observing such strange phenomenon?

Kalok

[jbednar]

Orientation selectivity is on an arbitrary scale, fundamentally, because it depends tightly on the number and type of patterns used -- e.g. responding to 1 out of 100 patterns tested is much more selective than 1 out of 4 patterns tested -- and thus it's only really meaningful as a comparison or when normalized across a set of measurements. In this case, the gcal.ty file sets topo.analysis.featureresponses.FeatureMaps.selectivity_multiplier=2.0, which gives a range 0 to 2.0. This value gives reasonable combined orientation preference/selectivity plots for GCAl, so that they are not too dark to see on average, but you can set it to whatever value you like.

The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.