IMPROVEMENT OF THE AUTOMATED NLP SYSTEM AS A FACTOR IN IMPROVING THE QUALITY OF MARKETING STRATEGY FORMATION

Abstract

Along with the rapid growth of the Internet of Things and edge computing, collaborative training of machine learning models on resource-constrained edge devices, while ensuring user data privacy, has become a key and challenging research challenge. The presented paper aims to fill this research gap by designing, implementing, and evaluating a synergistic and optimized federated learning infrastructure with differential privacy called PriFed-IoT, specifically designed for Internet of Things edge computing scenarios. The novelty of this work lies in the creation of a system of several modules working together, rather than a simple combination of techniques. The main idea is to use adaptive differential privacy to create an environment with a higher signal-to-noise ratio for the clustering algorithm in the late stages of training, allowing for more accurate separation of clients. In order to test the efficiency of the PriFed-IoT infrastructure, a series of complex simulation experiments were developed and conducted on a standard CIFAR-10 dataset, simulating different degrees of data heterogeneity. The study successfully offers a comprehensive solution, the experimental results convincingly prove the advantages of PriFed-IoT infrastructure in balancing privacy protection, model utility, and system efficiency, providing a valuable theoretical framework and technical implementation for building secure, efficient, and reliable intelligent applications at the edge of the Internet of Things.