Smart Classrooms Aided by Deep Neural Networks Inference on Mobile Devices

Machine Learning over edge computing devices is levering up embedded and IoT intelligence and is expected to grow even more. Today, Machine Learning applications are mainly driven by Cloud Computing, but a recent trend towards on-device ML execution is taken over. We can expect that in coming years there will be smart homes and buildings populated by much more smart devices able to assist human activities more naturally. We present a work in progress, a mobile edge computing prototype built to explore and validate how ready are smartphones for on-device execution of deep neural networks (DNNs) using as a test bed a light control of a smart classroom via object recognition using three pre-trained non-optimized DNN models embedded in one mobile app. The prototype was successfully accomplished for recognition and control tasks using only smartphone's CPU/GPU processing units to run DNN models under 670ms. Finally, on-device DNN inference performance is discussed and some future work issues are presented.