An Optimized Data Architecture for Smart Advertising based on Federated Learning

This paper presents a novel data architecture designed specifically for smart advertising applications, focusing on the efficient flow of data while ensuring high security and data integrity. The research leverages federated learning as a black box approach, emphasizing data authenticity, privacy, and the early removal of unwanted or fake data. The proposed data model adopts a semi-random role assignment method based on various criteria to collect and aggregate data. The architecture consists of model nodes, data nodes, and validator nodes, assigned to each node based on factors such as computing power, connection quality, and historical performance. By selecting only a fraction of nodes for modeling and validation tasks, the proposed architecture optimizes resource consumption and minimizes data loss. The AROUND social network has been used as a case study to demonstrate the effectiveness of the proposed data architecture. Both, the simulation and practical implementations demonstrate that it can reduce network traffic and average CPU usage by more than 50% when the number of users increased 20 times, while maintaining the performance of the recommender model.