Runtime Management of Artificial Intelligence Applications for Smart Eyewears

Artificial Intelligence (AI) applications are gaining popularity as they seamlessly integrate into end-user devices, enhancing the quality of life. In recent years, there has been a growing focus on designing Smart Eye-Wear (SEW) that can optimize user experiences based on specific usage domains. However, SEWs face limitations in computational capacity and battery life. This paper investigates SEW and proposes an algorithm to minimize energy consumption and 5G connection costs while ensuring high Quality-of-Experience. To achieve this, a management software, based on Q-learning, offloads some Deep Neural Network (DNN) computations to the user's smartphone and/or the cloud, leveraging the possibility to partition the DNNs. Performance evaluation considers variability in 5G and WiFi bandwidth as well as in the cloud latency. Results indicate execution time violations below 14%, demonstrating that the approach is promising for efficient resource allocation and user satisfaction.