Vieth and colleagues [1] have defined a plasticity rule that increases the synaptic weight by a fixed amount if the post-synaptic neuron fires within a fixed time interval after the pre-synaptic neuron, see their table S2.
NESTML generates incorrect code for this model with two types of errors:
If, following a pre-post spike pair that should trigger a weight increase, the postsynaptic neuron fires at least twice before the next pre-synaptic spike, plasticity will not occur.
If there are multiple pre-post spike pairs separated by min_delay intervals, plasticity does not occur.
The first issue seems to be cause by unsigned int num_transferred_variables = 0; in the generated set_spiketime() method. This variable should be 1.
The cause of the second issue not clear at present; it is of limited practical concern since the actual refractory time in the model by Vieth et al is more than 10 min delays, but it still seems to be a corner case that should be understood.
Attached below is an archive containing
NESTML model specifications for a parrot neuron designed to study plasticity and the synapse model by Vieth et al
a minimal reproducer for each of the two cases,
a script to build the necessary module.
[1] Vieth, M., Rahimi, A., Gorgan Mohammadi, A., Triesch, J., & Ganjtabesh, M. (2024). Accelerating spiking neural network simulations with PymoNNto and PymoNNtorch. Frontiers in Neuroinformatics, 18, 1331220. https://doi.org/10.3389/fninf.2024.1331220
Vieth and colleagues [1] have defined a plasticity rule that increases the synaptic weight by a fixed amount if the post-synaptic neuron fires within a fixed time interval after the pre-synaptic neuron, see their table S2.
NESTML generates incorrect code for this model with two types of errors:
The first issue seems to be cause by
unsigned int num_transferred_variables = 0;
in the generatedset_spiketime()
method. This variable should be 1.The cause of the second issue not clear at present; it is of limited practical concern since the actual refractory time in the model by Vieth et al is more than 10 min delays, but it still seems to be a corner case that should be understood.
Attached below is an archive containing
[1] Vieth, M., Rahimi, A., Gorgan Mohammadi, A., Triesch, J., & Ganjtabesh, M. (2024). Accelerating spiking neural network simulations with PymoNNto and PymoNNtorch. Frontiers in Neuroinformatics, 18, 1331220. https://doi.org/10.3389/fninf.2024.1331220