Technology for Selecting Deep Learning Models Based on Actualized User Experience Using the Auto-ML Approach

In the contemporary era of significant advancements in photo and video processing, object recognition and classification algorithms find extensive application in both personal and professional domains. However, the expansion of their usage areas brings an increase in factors influencing the diversity of computer vision tasks and the quality of image processing, including object dynamics and deformation within the frame. The growth in the number of neural network architectures complicates the selection of effective models to solve such tasks, posing new challenges to the developers of these algorithms. Our approach to addressing this issue involves creating our recommendation system ( https://saaresearch.github.io/ ), based on the logic of production rules formed from the analysis of previous research and scientific data. During our study, methods and technologies for creating recommendation systems, as well as methods for extracting statistical data and metadata from raster image sets, were analyzed. This work discusses the development of a framework that serves as a library of algorithms implemented to solve machine learning tasks. Special attention is given to the technology of selecting deep learning models based on updated user experience using the Auto-ML approach. Thus, the framework enables more efficient selection of deep learning models under various conditions of object detection tasks, including the original datasets of users, and facilitates the training of selected models in different modes (on CPU, GPU, multi-GPU) with specified parameters. The main goal of this work is to create a tool that automates the process of selecting optimal machine learning models, considering the user’s experience and preferences, thus significantly improving the efficiency and accuracy of model development in various applications. The results presented in this work demonstrate the high efficiency of the proposed technology, emphasizing the importance of an integrated approach to data analysis and machine learning in creating recommendation systems to solve a wide range of computer vision tasks, including those in industry.