Deploying Lightweight Neural Networks on Edge Devices Using Knowledge Transfer

When analysing the data right away with as little delay as possible, it becomes crucial to use lightweight neural networks on edge devices. IoT sensors, smartphones, and wearable devices invariably impose computational limitations and cannot support complex neural networks. This issue is solved by knowledge transfer techniques where the teacherstudent paradigms train smaller and less rigorous student models to perform like the more complex teacher models. This process achieves high accuracy while decreasing model size and computational requirement considerably. Real-life applications are self-driving cars making decisions quickly, home automation to control, and the constant monitoring of personal health using wearable devices that need to analyze physiological information on the go. A balanced design with performance and efficiency is made with aspects of knowledge transfer interwoven into these compact models. This paper discusses the application of LWNNS on edges, the real-time realizations of edges, and the issues and resolutions related to implementing scalable and efficient edges.