Zebra exploration to fine-tune parameters

[jeromelecoq]

The use of Zebra as a tool for system identifications of our neuronal population is incredibly exciting.
By combining all stimuli in a compact movie, we are potentially freeing a lot of time for the predictive learning part of our experiments.

Given our ambitions to understand how excitatory and inhibitory neurons work together, it is important we are able to characterize both populations effectively. A stimulus that only map pyramidal cell receptive field would miss the mark with a pan-neuronal line.

The initial Zebra efforts were focused in excitatory neurons, so it is likely that some parameters adjustments are necessary to strengthen inhibitory neurons excitation by Zebra.

Saskia DeVries @saskiad kindly agreed to share this analysis that she made. She used experiments that we released as part of the Visual Coding data.

This plot is very important. It shows that PV, SST and Pyramidal cells have quite different spatial and temporal frequency tuning.

[jeromelecoq]

Here is a direct screenshot of how Zebra looks on the UCL 2 screen set-up.

The UCL screen is 10 cm away and covers 180 of horizontal visual degrees. Note that the stimulus was not spherically wrapped on the UCL screen.

[jeromelecoq]

First in space, I explored how the xscale and yscale parameters influence spatial frequencies :

[jeromelecoq]

The equivalent UCL movie is at about 1.2-1.4 scale, given our distant screen. Note how spatial frequencies goes down quickly at 0.06 cpd right BEFORE the peak of SST cells.

So a better version of zebra would have a lower value of scale. A good trade-off would likely be scale at 1.0 for a screen 15 cm away such as our.

[jeromelecoq]

Now let's look at the temporal frequencies :

[jeromelecoq]

Here is the UCL value is tscale 50. Note that here we might actually be cutting also the responses of some excitatory neurons, based on our histogram. A better speed would cover this distribution more fully.
A good trade-off might be tscale 30

[jeromelecoq]

Based on this here is our candidate Zebra noise to display on the Allen Screen !

Here tscale is 30 and xcale=yscale=1.
We believe this should also recruit the inhibitory population.

allen_screen_tscale_30.mp4