Self Optimizing Neural Networks SONN-3 for Classification Tasks

The paper introduces new valuable improvements of a performance, a generalization ability and a topology optimization of the Self-Optimizing Neural Networks (SONNs). The described SONN-3 integrates the very effective solutions used in the SONN-2 together with the very effective ADFA algorithms for an automatic conversion of real input features into binary vectors. The integration provides not only a simple sum of valuable features of the both methods but it makes able to substantially improve a performance and generalization properties of these networks reducing SONN-3 topology sizes in comparison to SONN-2 topology sizes. The paper describes the construction of the SONN-3 and compares its performance with the SONN-2 and other AI computational methods applied to an exemplar classification task.