First clipped Gauss - last with homogeneous activation

This release is from the clipped Gauss branch where odors are not Gaussian binding profiles but clipped Gaussian, with tails set to proper 0. This allows to get rid of some artefacts that allowed very low binding OR types to dominate the response at very high concentrations when stronger binding ones already are suppressed / suppressed themselves (depending on inhibition pattern).
This version was used for the comparative trials on correlation with / without and covariance with / without self-inhibition.
While not written up in a proper report, the observations where that non-monotonic behaviour was possible, occurred for broad odors as a rule (no detailed investigation of whether there are other cases of non-monotonic behaviour where the odor is not broad yet done), and so far no case was found where the mixture of a monotonic strong odour and a non-monotonic odour would give the desired observation of close to linear superposition.
From here, next versions will add non-homogeneous activation rates for odors (k_2 + individual for each odor).

Rectified covariance instead of correlation

This is essentially as v1.3 but with rectified covariance for the inhibition strength. This inhibits only glomeruli that are suitably active and by simple rectification negative covariance leads to no inhibition.
However, we observe that essentially the behaviour is still overall the same with broad odours being non-monotonic and their inhibition dominating the mixture.
Note, this is still with self-inhibition set to 0.

Version underlying report3 - heterogeneous inhibition - 1st attempt

In this release I am preserving the model simulation in the state used for the results in report3. The main change to earlier versions was the introduction of 100 random odours and setting the local inhibition in the antennal lobe as a function of the correlations between glomerular activations. The correlations were translated (possibly not entirely on purpose) to the range of values k*[0.5, 1.5], where k is regulating the overall strength of inhibition. k was 3e-5*10^{x/2}. x \in {-2,-1,-0.5,0,1,2} (the values of x are ino and appear in the file names of results files).
As report3 is illustrating, these simulations were not quite fully satisfactory in that the mixture of two odours, one monotonic and one non-monotonic is still leading to a non-monotonic behaviour whenever the non-monotonic odour has more than negligible concentration. This contradicts my understanding that a strong monotonic odour response should continue unaffected (at least for some example) when mixed with the non-monotonic odour. Knowing that non-monotonicity seems to be based on broad activation of glomeruli and resulting broad and strong inhibition it seems unlikely that in this model a monotonic odour could be unaffected like this.

One odour - testing increasing breadth - final

This is the final version used to create report2.

One odour - testing increasing breadth

This version was used to test increasing breadth of odour profiles and systematically assessing whether the response increases monotonically with concentration (runs from 23 Feb 2021). This was specifically done using "exp2.py". The colormaps for maximum average PN firing as a function of the std of the Gaussian odour profile and the concentration are included in report2.

One narrow odour but more LNs

Similar to v1.0 but with a larger number of local neurons. Other parameters have been adjusted so that the same behaviour was achieved. This wasn't completely trivial as with the many LNs the connection scheme led to WTA behaviours of individual neurons/ glomeruli and no suppression at higher concentrations. What solved the problem was to make connections with a fixed pre-synaptic number, so that neurons are excited more equally.

One narrow odour stimulation

This first version was shared with Albrecht Haase in a first report early January 2021.
Discussed 12 January 2021 - we are able to demonstrate that a single odour can evoke non-monotonic responses with respect to input concentration due to the balance of excitation and inhibition in the AL.
The model was essentially well received. Point of note:

• consider verification model with 600 ORNs/ type
• consider augmenting LNs to 25 per glomerulus to mimic assumed numbers in the bee AL
• future directions:
  -- investigate variation of the width of the odour response. Establish whether non-monotonic response is dependent on wide or narrow response profile
  -- investigate a set of odours; presented individually and show whether a LN-PN connection pattern can be found for which the majority of odours evoke monotonically increasing responses but a few (a single one?) not
  -- Experiment: Verify with EAG that the response to geosmin is monotonically increasing at the level of ORs/ ORNs