artificial neural networks

Самоорганізація в моделях обчислень нейронних мереж і декларативних програм

Порівнюються дві моделі обчислень – штучні нейронні мережі та декларативні програми, що побудовані на основі логіки предикатів. Пропонуються такі їх узагальнення, за яких процес обчислень зможе приводити до цілеспрямованих змін власної програми. Сформульовано принципи самоорганізації, за якими вказані зміни будуть не хаотичними, а визначеними в результаті пошуку. Для детального вивчення самоорганізації необхідне зближення та взаємне доповнення розглянутих моделей.

Simultaneous surrogate modeling and dimension reduction using unsupervised learning. Application to parametric wing shape optimization

This paper presents a machine-learning-based approach that enables simultaneous surrogate modeling and dimension reduction and applies it to aerodynamic parametric shape optimization.  Aerodynamic shape optimization is a crucial process in various industries, including aerospace, automotive, and renewable energy.  It involves iteratively improving the properties of a system by evaluating an objective function and driving its minimization or maximization using an optimization algorithm.  However, the evaluation of aerodynamic objective functions requires computationally

Robust shape optimization using artificial neural networks based surrogate modeling for an aircraft wing

Aerodynamic shape optimization is a very active area of research that faces the challenges of highly demanding Computational Fluid Dynamics (CFD) problems, optimization with Partial Differential Equations (PDEs) as constraints, and the appropriate treatment of uncertainties.  This includes the development of robust design methodologies that are computationally efficient while maintaining the desired level of accuracy in the optimization process.  This paper addresses aerodynamic shape optimization problems involving uncertain operating conditions.  After a review of pos

Backpropagation algorithm for complex neural networks

Розглянуто комплексні штучні нейронні мережі, функції активації яких є комп- лексними аналогами раціональної сигмоїди. Наведено алгоритм навчання цих мереж, заснований на методі зворотного поширення похибки.

Neural networks with complex weights and continuously differentiable activation function have been studied in the paper. Learning algorithm based on the backpropagation method for rational sigmoid function has been given in the paper.

Технологія нейрокомп’ютингу реального часу

Проаналізовано особливості апаратної реалізації штучних нейронних мереж, вибрано принципи побудови, визначено шляхи підвищення ефективності використання обладнання, розроблено методи синтезу та базові структури нейрокомп’ютерних систем реального часу.

Features of hardware representation of artificial neural networks were analyzed, principles of construction were chosen, ways of efficiency increase of equipment use were determined, methods of synthesis and base structures of the neural computing, , real-time systems were developed.

A hybrid model for predicting air quality combining Holt–Winters and Deep Learning Approaches: A novel method to identify ozone concentration peaks

Ozone (O$_3$) from the troposphere is one of the substances that has a strong effect on air pollution in the city of Tanger.  Prediction of this pollutant can have positive improvements in air quality.  This paper presents a new approach combining deep-learning algorithms and the Holt–Winters method in order to detect pollutant peaks and obtain a more accurate forecasting model.  Given that LSTM is an extremely powerful algorithm, we hybridized with the Holt–Winters method to enhance the model.  Making use of multiple accuracy metrics, the models' efficiency is investig

Intelligent system for analyzing battery charge consumption processes

The article develops an intelligent system of analysis and neural network forecasting of battery charge consumption for automated vehicles (AGVs). For this purpose, the types of AGV and the methods of effective forecasting of their battery charge consumption were analyzed. It is established that they are based on optimal robot control processes; application of technologies to increase capacity and extend service life.

Comparative analysis of the specialized software and hardware for deep learning algorithms

The automated translation, speech recognition and synthesis, object detection as well as emotion recognition are well known complex tasks that modern smartphone can solve. It became possible with intensive usage of algorithms of Artificial Intelligence and Machine Learning. Most popular now are implementations of deep neural networks and deep learning algorithms. Such algorithms are widely used in all verticals and need hardware accelerators as well as deep cooperation between both software and hardware parts.

System of Processing of Technological Information

The problems of information processing in solving the technological preparation of production were considered. For this purpose use the effective methods of multivariate statistical analysis and artificial neural networks. Compression algorithms in the original array of information by factor analysis methods, component analysis and multidimensional scaling, classification algorithms and pattern recognition methods of discriminate and cluster analysis, as well as algorithms for modeling of group account of arguments and artificial neural networks were implemented with software.