A Review on Machine Learning in IoT Devices
Deployment of machine learning and AI continue to rapidly expand across a range of applications. Billions of devices are getting interconnected using IoT and the number of connected devices continues to grow exponentially. In edge computing the computation and intelligence are being brought closer to the edge devices like the gateways. Recently there is a growing interest in running machine learning models in the low-end devices particularly in embedded devices like microcontrollers (MCUs). Unlike high performance CPUs and GPUs these are resource constrained devices in terms of memory and computing power. This new paradigm of bringing machine learning and embedded devices together is referred to as TinyML. In this paper we present the motivation behind this paradigm, the challenges, and approaches to overcoming these challenges, popular frameworks and future directions.
Keywords: Edge computing, IoT, microcontroller, machine learning, embedded system, tiny machine learning