Pervasive AI for IoT Applications: A Survey on Resource-Efficient Distributed Artificial Intelligence

Artificial intelligence (AI) has witnessed a substantial breakthrough in variety of Internet Things (IoT) applications and services, spanning from recommendation systems to robotics control military surveillance. This is driven by the easier access sensory data enormous scale pervasive/ubiquitous devices that generate zettabytes (ZB) real-time streams. Designing accurate models using such streams, predict future insights revolutionize decision-taking process, inaugurates pervasive as worthy paradigm for better quality-of-life. The confluence computing artificial intelligence, Pervasive AI, expanded role ubiquitous IoT mainly collection executing distributed computations with promising alternative centralized learning, presenting various challenges. In this context, wise cooperation resource scheduling should be envisaged among (e.g., smartphones, smart vehicles) infrastructure (e.g. edge nodes, base stations) avoid communication computation overheads ensure maximum performance. paper, we conduct comprehensive survey recent techniques developed overcome these challenges AI systems. Specifically, first present an overview computing, its architecture, intersection intelligence. We then review background, performance metrics particularly Deep Learning (DL) online running system. Next, provide deep literature communication-efficient techniques, both algorithmic system perspectives, inference, training learning tasks across combination devices, cloud servers. Finally, discuss our vision research