Adaptive synaptogenesis constructs networks which allocate network resources by category frequency

Demonstrates the effectiveness of two adaptive processes in constructing simple, feedforward networks which allocate the resources of the output layer based on the frequencies of the categories that compose the input environment. Specifically, the adaptive processes build networks which allocate more output layer resources to categories that appear more frequently in the input. In turn, less frequently appearing input categories are allocated fewer output resources. The two adaptive processes, synaptogenesis and associative modification, build a network connectivity in initially unconnected networks. The first process, synaptogenesis, creates new synaptic connections; the second process, associative synaptic modification, modifies the strength of existing synaptic connections. Because these processes operate in a unsupervised fashion using only information available locally at the neurons and synapses, they provide a biologically plausible model for the allocation of neural resources