Cognitive Popularity Based AI Service Sharing for Software-Defined Information-Centric Networks

Abstract

As an important architecture of next-generation network, Software-Defined Information-Centric Networking (SD-ICN) enables flexible and fast content sharing in beyond the fifth-generation (B5G). The clear advantages of SD-ICN in fast and efficient content distribution and flexible control make it a perfect platform for solving the rapid sharing and cognitive caching of AI services, including data samples sharing and pre-trained models transferring. With the explosive growth of decentralized artificial intelligence (AI) services, the training and sharing efficiency of edge AI is affected. Various applications usually request the same AI samples and training models, but the efficient and cognitive sharing of AI services remain unsolved. To address these issues, we propose a cognitive popularity-based AI service distribution architecture based on SD-ICN. First, an SD-ICN enabled edge training scheme is proposed to generate accurate AI service models over decentralized big data samples. Second, Pure Birth Process (PBP) and error correction-based AI service caching and distribution schemes are proposed, which provides user request-oriented cognitive popularity model for caching and distribution optimization. Simulation results indicate the superiority of the proposed architecture, and the proposed cognitive SD-ICN scheme has 62.11% improved to the conventional methods.