Generating complex connectivity structures for large-scale neural models

Martin Huelse: Generating complex connectivity structures for large-scale neural models. In: V. Kurkova, R. Neruda, and J. Koutnik (Eds.): ICANN 2008, Part II, LNCS 5164, pp. 849?858, 2008. Sponsorship: EPSRC

Biological neural systems and the majority of other real-world networks have topologies significant different from fully or randomly connected structures, which are frequently applied for the definition of artificial neural networks (ANN). In this work we introduce a deterministic process generating strongly connected directed graphs of fractal dimension having connectivity structures very distinct compared with random or fully connected graphs. A sufficient criterion for the generation of strongly connected directed graphs is given and we indicate how the degree-distribution is determined. This allows a targeted generation of strongly connected directed graphs. Two methods for transforming directed graphs into ANN are introduced. A discussion on the importance of strongly connected digraphs and their fractal dimension in the context of artificial adaptive neural systems concludes this work.

Non peer reviewed