To obtain access to full text of journal and articles you must register!
- Article name
- The architecture of promising neurons for processing biometric data with high mutual correlation dependence (Review)
- Authors
- Sulavko A. E., , sulavich@mail.ru, Omsk State Technical University, Omsk, Russia
- Keywords
- statistical functionals / keyboard handwriting / biometric feature / voice parameters / signature features / wide neural networks / random number distribution law
- Year
- 2018 Issue 1 Pages 35 - 48
- Code EDN
- Code DOI
- Abstract
- The paper considers and summarizes various variants of constructing neurons in "wide" networks for pattern recognition problems in the presence of a small amount of training sample. Some of functionals that can be used as a basis for the formation of promising neurons able to work effectively in feature space with a high mutual correlation dependence are described. Stable statistical regularities between correlation dependence of features, their informativeness and parameters of some functionals based on criteria for testing hypotheses about the law of distribution of a random variable are found. The found patterns are confirmed by calculations for real biometric features (voice, keyboard handwriting, handwritten image, face). A transformation is proposed that allows to modify the differential functional on the basis of the Gini criterion for effective work with strongly dependent biometric features.
- Text
- To obtain access to full text of journal and articles you must register!
- Buy
