Show Reference: "Neural blackboard architectures of combinatorial structures in cognition"

See the CiteULike entry for more info, PDF links, BibTex etc.

Combinatorial or compositional codes are more efficient than codes based on grandmother-cells in terms of the number of cells to code an instance. They also generalize better.

A network with Hebbian and anti-Hebbian learning can produce a sparse code. Excitatory connections from input to output are learned Hebbian while inhibition between output neurons are learned anti-Hebbian.