The object of research is the process of classifying objects in images. The quality of classification refers to the ratio of correctly recognized objects to the number of images. One of the options for improving the quality of classification is to increase the depth of neural networks used. The main difficulties along the way are the difficulty of training such neural networks and a large amount of computing that makes it difficult to use them on conventional computers in real time. An alternative way to improve the quality of classification is to increase the width of the neural networks used, by constructing ensemble classifiers with staking. However, they require the use of classifiers at the first stage with different structured processing of input images, characterized by high quality classification and relatively low volume of calculations. The number of known such architectures is limited. Therefore, the problem arises of increasing the number of classifiers at the first stage of the ensemble classifier by modifying known architectures. It is proposed to use blocks of rotation of images at different angles relative to the center of the image. It is shown that as a result of structured image processing by the starting classifier, processing of rotated image leads to redistribution of errors on image set. This effect makes it possible to increase the number of classifiers in the first stage of the ensemble classifier. Numerical experiments have shown that adding two analogs of the MLP-Mixer algorithm to known configurations of ensemble classifiers reduced the error from 1 to 11 % when working with the CIFAR-10 dataset. Similarly, for CCT, the error reduction was between 2.1 and 10 %. In addition, it has been shown that increasing the MLP-Mixer configuration in width gives better results than increasing in depth. A prerequisite for the success of using the proposed approach in practice is the structured image processing by the starting classifier

Досліджено можливість швидкого обчислення колек тивних експертних оцінок медіанного типу. Незважаючи на широке застосування для розрахунку колективних екс пертних оцінок медіан КеменіСнелла і КукаСейфорда, недостатньо досліджені можливості скорочення часу обчислення медіанного консенсусного ранжирування шля хом застосуванням задачі про призначення і відомих алгоритмів її рішення. На відміну від більшості відомих методів пропонований в статті метод не є наближеним і зберігає вихідну медіанну аксіоматику Кемені. Досліджено можливість розрахунку медіан Кемені Снелла і КукаСейфорда із застосуванням задачі про призначення методами комп'ютерного експерименту. Оцінено час рахунку медіанних ранжирувань чотирма різними алгоритмами рішення задачі про призначення. Встановлено, що запропонованим методом при помір ній кількості альтернатив (n<50) медіанні ранжируван ня розраховуються в режимі часу, близькому до реально го. Показано також, що на відміну від інших методів час обчислення медіанних ранжирувань за допомогою задачі про призначення не залежить від ступеня узгодженості індивідуальних експертних ранжирувань. Отримані результати корисні для практичного засто сування дослідженої процедури в мережевих експертних системах. У таких системах час обчислення консенсус ного ранжирування має бути близьким до реального. Крім того, в мережевих системах експертизи за рахунок випад кового комплектування колективу експертів можливий низький рівень узгодженості індивідуальних ранжируван ня. Для дослідженої процедури це не впливає на трива лість розрахунку. Це дозволяє рекомендувати розробле ну обчислювальну процедуру швидкого пошуку медіанних консенсусних ранжирувань за КеменіСнеллом і Куком Сейфордом для практичного застосування в системах колективної мережевої експертизи Ключові слова: колективне експертне оцінювання, ме діанні консенсусні ранжирування, задача про призначення UDC 004.9

No abstract

The object of research is the process of forming a collective expert assessment with increased reliability in making management decisions in business structures by an expanded team of experts. One of the most problematic places in the expert assessment of management decisions is the complexity of forming a competent expert team and the rather high cost of the expertise. In recent years, there has been a tendency for expert assessment with an expanded team of experts. In this case, not only professional experts are involved in the examination, but also all persons wishing to take part in solving the problem. In this case, the reliability of the examination raises doubts. In connection with the participation in expert assessment of persons who do not have experience in expert work, a wide range of expert assessments is possible. The analysis of the current state of the methods of expert assessment in business is carried out. It has been established that the Delphi method, which was most used until recently, does not meet modern requirements. More progressive methods are based on mathematical consensus theory. Consensus is understood as the degree of correlation of individual expert assessments performed in rank scales. In the course of the study, formalized mathematical approaches to the organization of collective expertise were used. A method for processing the results of an examination with an expanded composition of experts was developed. The developed methodology is focused on identifying experts with insufficient qualifications. The methodology allows for a step-by-step assessment of the reliability of the collective expert decision by assessing the Kendall concordance coefficient. It is shown that the phased exclusion of assessments by experts with insufficient qualifications allows increasing the level of consensus, the quality and reliability of the collective expert assessment. The developed methodology has been tested in a really functioning enterprise to make a decision on the exit strategy of the enterprise from their crisis. The use of the developed methodology has made it possible to significantly increase the reliability of the examination results, assessed by the concordance coefficient. The results are useful for practical application in business structures when conducting expert examinations involving a wide range of participants.

Expert evaluation and reasonable selection of digital components in the microelectronic market is a complex and responsible task. For its solution, the known methods of carrying out expert estimations do not fit fully in connection with the laboriousness of the results processing. The development of an expert choice method for digital components that allows you to quickly obtain a generalized collective expert evaluation (CEE), evaluate the consistency of expert opinions and make informed decisions is a quite actually. The goal of the study is to develop a method for forming a voucher for the selection of digital components of industrial automation systems based on the Markov chain and its verification in the real practical situation. A method is proposed for CEE forming for complex components of automation systems based on the Markov model. When aggregating expert preferences, each alternative is represented as a state of the Markov chain. Next, for the vertices of a Markov graph, the Copeland number is calculated, equal to the difference between the number of arcs entering and leaving the vertex. In collective ranking, alternatives are arranged in descending Copeland numbers. The developed method has a high speed in comparison with the known analogs. The correctness of the proposed method, its efficiency and speed has been confirmed by real expertise and in the process of computer modeling. The executed researches showed that the developed method for the collective expert evaluation forming works 80-200 times faster than the method based on the median Kemeni. The practical significance of the proposed method has been demonstrated on the real expertise carried out at the enterprise «Krioprom» (Odessa, Ukraine) when purchasing a batch of programmable logic microcontrollers within the large-scale project framework for cleaning units automation of industrial air-separation plants.