Scalable and Energy-Efficient Deep Learning for Distributed AIoT Applications Using Modular Cognitive IoT Hardware

In this scientific paper, we present a novel approach to develop energy-efficient Deep Learning models for distributed AIoT applications. Our approach considers the optimization of algorithms, while also addressing safety and security challenges that arise in such systems. We propose a modular and scalable cognitive IoT hardware platform that leverages microserver technology, allowing users to customize hardware configurations to suit a broad range of applications. We provide a comprehensive design flow for developing Next-Generation IoT devices that can collaboratively solve complex Deep Learning applications across distributed systems. Our methods have been thoroughly tested on diverse use-cases, ranging from Smart Home to Automotive and Industrial IoT appliances. Our results demonstrate the effectiveness of our approach in significantly reducing energy consumption while maintaining high performance in Deep Learning applications. Overall, this work contributes to advancing the development of energy-efficient and scalable Deep Learning for distributed AIoT applications, providing an important step towards enabling the deployment of intelligent systems in diverse real-world scenarios.