Layer

From Emergent

A layer contains Units (individual neuron-like processing elements), and provides some structuring to how computations are performed on those units, depending on the algorithm. Layers are used to organize Projections (connectivity), and so provide the basic structural skeleton of the Network.

In Backpropagation, the order of layers within the network determines the ordering of activation and error flow (forward and backward).

In Leabra, layers play a large role, determining the scope of direct inhibitory competition among the units.

Layers can be subdivided into Unit Groups which in Leabra can provide more fine-grained levels of inhibitory competition, and can also be used to organize patterns of connectivity between layers.