Conditions of the correctness for the algebra of estimates calculation algorithms with μ-operators over a set of binary-data recognition problems
A. E. Dyusembaev, Mikhail Grishko · Pattern Recognition and Image Analysis · 2017
This paper is aimed to show that specialized neural networks can be useful for finding exact solutions of the recognition problems involving binary data. For this purpose, as an original subclass, we take the subclass of estimates calculation algorithms (ECAs) in which all algorithms correspond to three-level multilayer neural networks (μ-blocks). The correctness conditions are defined that allow a correct algorithm to be constructed in the algebra over this ECA subclass for each Ω-regular recognition problem. This approach does not require additional constraints, except for a trivial one, on the system of ECA support sets.