As a contribution to increasing the range of ideas on architecture and process for incorporatin in ANNs a new theory is outlined of the emergence of parcellation of the cerebral cortex on an evolutionary time scele. Slow learning and accelerated evolution, involving a form of inheritance of acquired characteristics, are assigned fundamental roles in the creation of functionally tilted, local cortical area architectures. Within each generation a cycle of neuron -> astrocyte -> neuron interaction produces a web of associated astrocytes defining local neural areas consistently engaging in integrated subsymbolic processing. Effects of intra-generational experience enter the germ line via processes involving astrocytes, epithelial cells, lymphocytes and RNA retroviruses. Potential application of the theory is explored in evolutionary programming aimed at constructing a generalisable, recurrent network induction algorithm.
展开▼