This is the readme for the models associated with the paper:
Top-down inputs drive neuronal network rewiring and context-enhanced
sensory processing in olfaction, Wayne Adams, James N. Graham,
Xuchen Han, Hermann Riecke, PLoS Computational Biology (2018)
This code was contributed by Hermann Riecke.
To run the matlab code there needs to be a file fignum.txt in the same directory as the code bulbpiri.m and
a subdirectory called Saved Research. During the initialization of the code it makes a new subdirectory within
Saved Research that is labeled with a consecutive number based on the value in fignum.txt and the current date
and time. All figures are saved there.Note: in the uploaded code there may be slight variations of the parameters for Figs. 3, 5, 7 compared to those used
in the publication.
The different cases corresponding to the various figures are controled with the variables learning case and probe case.
The correspondence with the key figures is as follows
* Fig.3: learning case=8
* Fig.5: learning case=6
* Fig.6: learning case=13
* Fig.7: learning case=7
In the uploaded code the parameter number GC aim final was set to 500 to speed up the computations. To
reduce the fluctutions in the connectivities that number should be raised to the values given in the publication. The
temporal evolution is relatively slow. To get a reasonable impression of the resulting connectivities and activities
one has to run the code for at least 40,000 steps. In the code this corresponds to setting outer param step=400.
Brief description of the output figures:
* Figure 3 shows the activity of the GC with rectifier (left panel) for the training stimuli. The second and third
panel shows the contributions to the GC activity from bulbar and cortical input, respectively. Panel 4 shows
the pattern correlation between the GC activities for the training stimuli.
* Figure 5 shows the cortical connectivity Wcc as well as the activity of CC without and with sigmoid.
* Figure 16 shows the effective disynaptic inhibition WMMand WMC.
* Figure 17 shows the CC and MC activities for the training stimuli.
* Figure 18: The top 2 panels show the correlations between stimulus pair 1, 2 and stimulus pair 3, 4 as well as
the corresponding Fisher discriminants as a function of time. Panel 3 shows the contributions of each MC to
the Fisher discriminants. Panels 4, 5 show the MC acvtivity without and with rectifier.
* Figure 19: like Figure 18, but without cortical input.
* Figure 80: Top left and bottom left panels show the MC and CC activities for the probe stimuli shown in the
top right panel. For each stimulus the activities are shown without context and with contexts 1 and 2. The
bottom right panel shows the same MC activities but as line plots rather than heat maps.
* Figure 81 shows the results corresponding to Figure 80 in the absence of cortical inputs.
* Figure 1018 shows the temporal evolution of the Fisher discriminant for random read-out as a function of
* Figure 1019 shows the results corresponding to Figure 1018 in the absence of cortical inputs.