This invention is a system and iterative non-learning method to determineoptimal artificialneural network node and layer count, edge connection structure and transferfunction for anartificial neural network. Optimality is indicated by the learning effort forthe network beingminimum and the generalization of the artificial neural network on provideddata beingmaximum. A control and display subsystem receives a count of input and outputexternalinterface nodes and associated node names from a user. Said subsystem alsoaccepts end-conditions for the training of a series of artificial neural networks andestablishes, togetherwith a data delivery agent and data mapping agent, a relationship betweenvariables of a dataset and input and output network nodes. A network configuration agent andconfigurationagent controller create a series of artificial neural networks, each networkin a series having adifferent internal nodal structure or transfer function than others in theseries. A trainingagent trains each configured artificial neural network over an epoch oftraining, for eachmodified artificial neural network in a series. A data-logger records thetraining progress.An analyzer computes the improvement or reduction of training efficiency andability of aparticular network structure to generalize on a provided data set, compared toa previousstructure. The control and display subsystem, analyzer and configuration andtrainingcontroller subsequently determine a probable best structure of artificialneural network for asubsequent iteration of network creation and training testing.
展开▼