Sparse Connection


Sparse connectivity gets its name from the concept of a sparse matrix i.e. a matrix primarily populated by zeros. Likewise, since a given set of connections is generally represented as a "weight matrix" this corresponds to there being a relatively small number of connections between two sets of neurons. Specifically, the sparse connection panel allows users to specify how well connected two sets of neurons are in terms of a percentage of the total possible connections (0% being no connections and 100% being the same as connecting every source neuron to every target neuron). Technically, this means that sparse is capable of creating so called "dense" connectivity, since any percent is possible (so long as there are enough neurons). It should be noted that sparse connects sets of neurons randomly (i.e. if one has 4 source neurons and 4 target neurons and desires 50% connectivity 8 synapses will be generated connecting random source neurons to random target neurons), and futhermore it is impossible to designate or guarantee than any particular connection will be generated. In the aforementioned example there is no guarantee that there will be a connection between say source neuron 2 and target neuron 3, unless the percent connectivity is set to 100%. Sparse connection does guarantee that the number of synapses generated will be the closest possible approximation of the percent connectivity specified (since any percentage of possible connections must be an integer value). Parameters relating to inhibitory and excitatory attributes of these connections can be adjusted from the sparse panel via the excitatory/inhibitory properties sub-panel.