ATELIER: Service Tailored and Limited-Trust Network Analytics Using Cooperative Learning

Abstract

The current trends in mobile network architectural design are moving toward the adoption of open interfaces that allow data exchange among different stakeholders in the network. This open circulation of data, happening across all network domains, including Access and Core, aims to improve the network operation through the usage of Artificial Intelligence (AI)-based solutions. This paper focuses on the interaction among Service Providers delivering applications to their customers using the infrastructure of Mobile Network Operators. In this scenario, it is paramount that such interactions occur with limited trust, as network operators and service providers may be competitors in the market, therefore avoiding the exchange of raw data and labels. In this work, we propose ATELIER, a deep learning solution for the provisioning of tailored network analytics from the network operator to the service providers in a limited trust fashion. Our design, which leverages similarity learning and reinforcement learning solutions, demonstrates how it can improve the analytics for a multimedia streaming service under various service configuration parameters, such as the video type and the desired Quality of Experience (QoE) level for the end users, achieving a performance increase of up to 37.7% compared to other methods, while also doubling the precision.

Keywords

• Zero-trust networks
• network analytics
• cooperative learning