A Randomised Distributed Primer for the Updating Control of Anonymous ANNs

The computational potentiality of ultra-large numbers of processing units working in parallel may be nullified if proper co-ordination mechanisms do not exist. Surprisingly, most simulations and implementations of ANN models under discrete time (digital-based hardware and software in conventional and parallel computers) do not necessarily have built-in distributed forms of controlling the updating of their neuronal units. In general, these simulation/implementation strategies are tailored to particular classes or topologies of neural paradigms and do not present explicit scalability, i.e. whether or not the strategy is prepared to cope with leaps in orders of magnitude in the number of neural units. In this sense, the implications of associating a distinct identification to each processing unit are definitely very strong if a massive number of these units are considered. Differentiated by this critical aspect, anonymous distributed mechanisms have also an intrinsic biological plausibility appeal in the context of ultra-large ANNs.