On Brain-inspired Hierarchical Network Topologies

In this paper our aim is to identify layered hierarchical generic network topologies which could closely mimic brain's connectivity. Recent analyses have compared the brain's connectivity (based both on a cortical-equivalent Rent's rule and on neurological data) with well-known network topologies used in supercomputers and massively parallel computers (using two different interpretations of Rent's rule). These have revealed that all the well-known computer network topologies fall short of being strong contenders for mimicking the brain's connectivity. That is why in this paper we perform a high-level analysis of two-layer hierarchical generic networks. The range of granularities (i.e., number of gates/cores/neurons) as well as the fan-ins and the particular combinations of the two generic networks which would make such a mimicking achievable are identified and discussed.