Constructive Learning Neural Network Algorithms in Template Classification

Abstract

Constructive learning algorithms are considered as an interesting method for
constructing growth of near-least neural network architecture. They help to maintain
single-user request and come over with our inappropriate choices of network topology in
the algorithms. This includes searching for proper weights in our previously-fixed network
architectures. Some of these algorithms, convergent to zero classification errors, are shown
in such applications that include acquisition of a binary to binary map.
In this study, two constructive learning algorithms are presented named MPyramid-real
and MTiling-real, which develop Pyramid and Tiling algorithms for Real to M-ory maps,
respectively. In the current study, the convergence of algorithms, their applicability, and
how mixing of a local pruning stage may delete some additional neurons from MTilingReal network have been investigated.