ABSTRACT
Cyber-physical systems (CPSs) and the Internet of Things (IoT) are complementary paradigms integrating digital capabilities. The CPS is more oriented for vertical integration, enabling communication between physical systems, computation, networking, and software. IoT, however, emphasizes connecting things for horizontal applications today to lead to the Internet of Everything tomorrow. Smartness and data explosion are two essential attributes of IoT that have been highlighted in recent definitions. Collecting, analyzing, and controlling the data and the computation is where the real power of both the paradigms lie. Satisfactorily addressing both the issues is crucial for the development of IoT as well as CPS. Due to deep learning's success in many IoT applications, it has become evident that these models will play a crucial role in establishing the CPS. However, the models’ complexity, including their size, resource demand and training procedures, necessitates a comprehensive investigation for successfully establishing such models over CPS architectures. This chapter presents a theoretical framework for the distribution of deep neural networks over three-tier CPS architecture. Actuators, embedded systems that capture data, are closest to the physical world and form the architecture's first layer. The second layer consists of nodes with intermediate resource handling capabilities. The third layer is resource-rich and hosts most of the deep learning tasks. Further, a brief insight into current challenges and future work will also be provided.
