Supporting Internet of Things Devices with DNNS
Vaishali Singh, Tony Aby Varkey M, S Mohanraj · 2023
Deep neural network powered smart metres have become revolutionary technologies in several industrial areas as a result of the current upsurge in IoT and DNNs. nevertheless, dealing with information from IoT devices using DNNs results in substantial delay and usage of energy due to their complexity, which is characterised by a large number of parameters and data-processing procedures. Methodologies have been created to solve these issues and allow real-time DNN deployments on devices with restricted resources. This work provides a thorough analysis of various methods, focusing on the trade-off between simulation dimension, preciseness of classification, processing velocity, and fuel economy. A brief introduction of DNNs sets the scene for the evaluation at the outset. The accessible tools for the installation of DNNs using low-resource hardware devices are then covered in detail. The topic then shifts into dataflow mapping techniques and memory tree architectures, which are crucial for maximising DNN adoption on limited technology. Additionally, several model optimisation methods like compression and trimming are investigated in depth. This paper offers real-world case examples that demonstrate the viability of using DNNs in Internet of Things (IoT) applications in spite of theoretical reflections. Finally, the paper offers helpful advice for designing and delivery of future-oriented software and hardware systems geared for practical Internet of Things (IoT) apps by delving deeply into gaps in knowledge and future prospects.