Search academic texts

Registration Number

0216U007006, 0111U002092 , R & D reports

Title

Methods for neural network representation and processing of information in intellectual information technologies

popup.stage_title

Head

Rachkovskij D.A.,

Registration Date

15-03-2016

Organization

International Research and Training Center of Information Technologies and Systems UNESKO IIP

popup.description2

Randomized data presentation techniques based on neural network distributed representation of vector and complex structured data was developed. The methods of converting the original real vectors in the reduced dimensional real vectors as well as vectors of binary variable dimension was developed. Those methods allows to evaluate a measure of similarity, the difference of the input vectors computationally efficiently. The methods of stable solutions of discrete ill-posed inverse problem, which arises in the solution of many measuring data processing tasks, was developed. Those methods based on the use of a truncated singular value decomposition and the projection of a random matrix. The criteria of selection of the model to ensure the restoration of the initial vector with the error signal, which is close to the minimum possible, was established. The methods of distributed representations in the form of binary vectors of fixed dimension structured data of different types, such as character sequences and complex hierarchical structure in the form of labeled directed acyclic graphs with vertices - objects and relationships of different levels of the hierarchy, was developed. The approach to the extraction of relationships between objects of the texts using the distributed representation structures is proposed. A method of using distributed views and information on the sequence of words in the texts for the recognition of named entities is developed. As a result of research on the topic the theoretical basis of neural network randomized distributed representation and processing of data for intelligent information technology was developed. The developed presentation and methods of their processing allow an expansion of the range of problems solved by reasoning on the basis of examples, increasing the efficiency of their solution, as well as the ability to create new intelligent information technologies based on simulation elements of thinking and the structural and functional organization of the brain mechanisms.

Product Description

popup.authors

Гольцев Олександр Дмитрович
Лукович Володимир Володимирович
Ревунова Елена Георгиевна
Сліпченко Сергій Віталійович
Соколов Артем Михайлович

popup.nrat_date

2020-04-02