MACHINE LEARNING ON MOBILE DEVICESIMPLEMENTING

Abstract

The article reveals the architectural components of the implementation of machine learning on mobile devices. The main advantages of using deep learning and inference on a mobile device are highlighted: bandwidth savings, reduced cost of cloud computing resources, fast response time, mobile computing stores sensory data on the local device, which significantly improves the confidentiality of user data. It is emphasized that today there are five architectures that are commonly used to implement machine learning on mobile devices. At the same time, it is emphasized that each individual architecture, its fundamental basis depends on the details of the scenario, such as the specific requirements of the mobile application, the complexity of the model, the amount of data and so on. The first architecture is non-learning data output, based on the fact that the mobile application sends a request to the cloud through the application programming interface along with the new data, and the service returns the forecast. The second architecture is data output and cloud learning, based on the principle of the previous model, the only difference being that service providers give mobile device developers the ability to learn data and create their own unique models using the cloud service. The third architecture is output to devices with pre-trained models, the principle of implementation is based on the fact that the pre-trained model is loaded into the mobile application to make output, the mobile application runs all output calculations locally on the device. The fourth architecture is output and learning on the device, the program can constantly learn from the data and user behavior, and therefore, constantly update models and improve performance for this user. The fifth architecture is a hybrid architecture, based on the principle that the basic model is trained in the cloud using a large standard set of data or a large set of all data used by users. It is emphasized that today, the easiest way to include machine learning in a mobile application is to use a cloud service that covers the functionality of both components.