A Distributed Federated Learning Method Using Digital Signage in Tourist Destinations

Abstract

In recent years, AI-powered services have been proposed in the tourism sector. This paper proposes an improved decentralized federated learning method that utilizes digital signage placed at various tourist spots for improving model accuracy. In the proposed method, digital signage stores multiple models received from nearby mobile devices and integrates them to construct a more accurate model. Evaluation results show that the best models built on each mobile device using the proposed method achieves an average accuracy of 0.712, improving by 0.078 compared to conventional methods using only opportunistic communications between mobile devices. Additionally, the number of communications during training in the proposed method was 6.79 per mobile device, reducing by 15.9% compared to the method that utilizes both P2P communications and communications with digital signage.