yes. And of course t_syn of 70ms smooths the PSPs to such an extent that the individual PSPs are essentially undetectable.

OK, but the poor performance with high numbers of input spike trains is simply a function of the high output rate! (and lack of windows for detecting PSPs etc)

As mentioned before, plotting performance as a function of output rate and of total input spike rate could also give useful insights here.

Random thoughts:

• Is this anything to do with the SNR model? If I remember correctly, this is applied to the voltage range of Vmem, which will be a lot larger with spikes than without

• Also, as the SNR is so small, the VI signal appears to be mostly Gaussian noise with only a few inputs. So, with only a few inputs, there is little to cause a detectable deviation in the mean of the noise, but consecutive PSPs from multiple inputs will change the mean value upwards. Which presumably is what the STA is detecting.

The spike rate of each train and the number of trains should in principle have the same effect I would think. The important quantity will be the total number of incoming spikes (at least, for Poisson trains)

Great stuff - so now the next main step is to figure out how to summarise the results of the connection tests for a given simulation (i.e. quantify how well they do - for example, we might count all correct P-values at some arbitrary threshold for true positives and true negatives, and summarise using some signal detection measure).

And then show how that summary number depends on each of the properties here (e.g. recording duration, noise level), while holding the others fixed.

I'd still like to see a read-out of the spike-rate of the simulated neuron. This variable may turn out to be the key to successfully detecting connections - because it sets the limit of the time window in which a pre-synaptic neuron can elicit a detectable PSP without being masked by the post-synaptic neuron's spikes. i.e. the faster the post-synaptic neuron spikes, the harder detecting connections might be...

I could not see this in the code? I see a NaN - or am I looking in the wrong place?