APPLICATION OF THE KUBEFLOW TOOL FOR THE INTEGRATION OF MACHINE LEARNING AND ARTIFICIAL INTELLIGENCE IN UNMANNED AERIAL VEHICLE

At the current stage of information technology development, machine learning (ML) and artificial intelligence (AI) are becoming one of the main tools for solving complex applied problems in various fields of activity. Various processes and technologies are used for develop, test, and maintain the infrastructure of the data system. The application of tools for the integration of ML and AI in the management of unmanned aerial vehicles (UAVs) is especially relevant today. An overview of the ML concept and processes (Machine Learning and Operation, MLOps) was made, which is a set of techniques for implementation and automatic continuous integration, as well as delivery to the product environment and model learning. The concept of MLOps is considered in terms of Kubeflow tools, which work on the Kubernetes platform. The possibilities of using modern MLOps solutions to improve the development processes of ML information systems have investigated. An AI-based information system with the possibility of continuous learning has designed. The concept of using the MLOps pipeline to solve the applied problem of classifying objects from the video of reconnaissance UAVs was presented. The results of the operation of the model in the Kubeflow arsenal have been checked using such improvement factors as: speed of development, implementation of changes, reduction of time to search for problems, recovery after global interruptions, reduction of the number of errors in the model. A publicly available model was deployed in a Kubeflow cluster using the Seldon Core Serving application manifest for practical analysis. The conducted research showed that Kubeflow consists of a set of various open source components that have a high level of integration with each other through the Kubernetes platform. At the same time, Kubeflow uses the Kubernetes pattern of operators for ML objects extremely efficiently. It has shown that writing model code is a small part of ML tasks, which affects the need for automation. The concept of a full-fledged information solution based on the continuous integration pipeline, which is the foundation of the implementation of the concept of continuous learning, has formed. Representing abstractions in the form of separate platform resources allows you to reduce the entry threshold for the end user.