In this paper, we present the result of the research related to the possibility to measure numerically the explainability of already trained CNN model using methods based on the perturbations of input image. The level of explainability is measured as the average IOU over the binarized classification explanations provided by RISE, Grad-CAM, RD methods, and ground-truth labels of the dataset. The evaluation of explainability is performed for five networks having different architectures and accuracies and trained to solve the same cat/dog classification problem for Oxford-IIIT Pet Dataset with 37 classes. It has been shown that CNN with high accuracy does not necessarily have good IOU explainability. The decision about which model is better is split between different methods. The question about measuring the explainability becomes more acute when ground-truth labels are missing for the dataset, so qualitative evaluations with IOU are not possible. For this case the proposed RDE value may be used as some indicator of the explainability of the model. It is also shown that the performance of RD is better compared to other methods.
The object of the study is evacuation parameters from institutions of higher education with inclusive education, in particular, speed, flow density and the proportion of participants using crutches and wheelchairs. The relevance of the work is confirmed by the rapid development of inclusive education in Ukraine and abroad. This requires adaptation of existing models for calculating the evacuation duration for use in educational institutions with inclusive education. A way to solve this problem is to present the speed of movement of people during evacuation as a function not only of the flow density, but also of the participants with reduced mobility percentage. In the work machine learning methods, in particular, a linear regression model and an artificial neural network are used to establish the relationship between movement speed, flow density, the proportion of participants using crutches and the proportion of participants moving in wheelchairs. It has been proven that the use of the obtained dependence makes it possible to increase the accuracy of calculating the duration of evacuation from educational institutions with inclusive education by 14
Prospective clinical studies have demonstrated increased detection rates of breast cancer using computer-aided diagnosis (CADx). CADx technology is designed to reduce observational errors made by physicians and reduce the number of false negatives when interpreting medical data. In this paper, we present the architecture of the computer-based breast cancer decision support system CADx, which allows for the construction of such a system. A feature of our proposed architecture is the creation of a corporate cloud for deploying a specialized neural network for CADx of primary breast cancer and hosting a non-relational factual database of patient research results, which can be used as SaaS for machine learning in computer diagnostic systems. Cloud technologies make it possible to use virtually unlimited memory resources and the speed of remote servers in SaaS mode, which requires the direct user only to have a fairly fast Internet connection and does not require a powerful computer or other equipment. The combination of machine learning methods with cloud service architecture creates a powerful new tool for effective diagnosis of breast cancer that can be used everywhere, and does not require significant costs for either deployment or operation. The database contains a heterogeneous vector of primary measurements (metadata, DICOM files, and other images and data) for each patient, which facilitates the construction of a neural network for tumor recognition and preliminary classification. We have populated the database with new region-specific data regarding the health of women in Ukraine under the severe stress caused by the ongoing war.
The problem of developing effective extrapolation methods is still relevant today, despite the existence of many known methods and approaches. In this paper, we consider the case of small samples of known data and the method of extrapolation based on the construction of the set of all possible polynomial forecasts. We present the results of the research concerning an algorithm for finding the optimal degree of the interpolation polynomial for solving the extrapolation problem. A number of important refinements of this algorithm are proposed. In particular, the procedure for building a forecast in the case when the convergence of a number of polynomial forecasts is non-monotonic has been clarified. We consider the case when the series of polynomial forecasts diverges. A detailed study of this case showed that an effective forecast can be constructed by averaging the first terms of a series of polynomial forecasts, provided there is no monotonicity. A stochastic case was also considered using the example of the Bitcoin exchange rate. Analysis of the polinomial predictive set showed that the criterion for choosing the number of consecutive values of a series of polynomial forecasts is the minimum deviation of the corresponding arithmetic mean to the last known value of the input time series. The proposed algorithms are important because they make it possible to build effective forecasts of various processes based on polynomials.
The review presents an analysis of current methods, models, and tools employed in developing and optimizing Internet of Things (IoT) systems. It meticulously examines a variety of approaches and software solutions, highlighting the rapid advancement in the automation field, enthusiastically embraced by both leading international IT corporations such as Google, Apple, and Xiaomi, and domestic companies like AJAX, which are gaining market popularity. The manuscript underscores the significance of IoT across various facets of modern life, including energy efficiency, security, and business process optimization, emphasizing the continual evolution of technologies that unlock new opportunities for users and enhance their quality of life. Additionally, it discusses the integration of IoT in different sectors, potentially exerting a substantial positive impact on society and the economy. Furthermore, the article offers a historical overview of IoT development, charting its evolution and key milestones. It also explores contemporary communication technologies foundational to IoT devices and outlines essential criteria for modern reliable systems, assessing them in terms of efficiency, scalability, and security. An analysis of several popular off-the-shelf system solutions available in the market is also provided, detailing their features, advantages, disadvantages, and unique attributes. This enables a deeper understanding of the diverse options available, assisting in selecting an appropriate system based on individual needs and budget. The article serves as a valuable resource for those seeking updated and comprehensive information on modern IoT systems and their implementation, stressing the problem’s relevance in today’s rapidly digitalizing world, where IoT technology plays a crucial role in enhancing operational efficiencies and living standards.
