Table 3.

Descriptions of perturbation models used to produce imperfect graph reconstructions from a synthesized ground truth network.

Error type	Perturbation model description
Synapse deletion	A specified percentage of synapses is randomly selected from the set of all existing synapses and deleted.
Synapse insertion	For each possible pair of cylindrical process segments (from different neurons), insert a synapse with probability p where p is pmax for inter-process distance less than d1, p is 0 for distance greater than d2, and p follows a linear decreasing curve in (d1, d2).
Neuron split	For each cylindrical process segment, split the neuron at the segment with probability p where p is pmax for process diameter less than d1, p is 0 for diameter greater than d2, and p follows a linear decreasing curve in (d1, d2).
Neuron merge	For each possible pair of cylindrical process segments (from different neurons), merge the neurons at the segments with probability p where p is pmax for inter-process distance less than d1, p is 0 for distance greater than d2, and p follows a linear decreasing curve in (d1, d2).