Energy-efficient AI at the Edge

The advances in artificial intelligence, especially convolutional neural networks (CNNs), over the past few years resulted in state-of-the-art solutions for many tasks, e.g. computer vision. As more and more intelligent applications rely on these methods, there is a growing interest in processing the data locally, at the place of the generation: the rise of intelligent edge computing will enable new possibilities in many areas, like smart manufacturing, autonomous systems, remote control. To increase the performance of the applications, the complexity of the neural networks also increases, leading to growing computational costs. As typical edge computing systems are power constrained, power efficiency is a key factor in designing them. In this paper, we evaluate the performance and power efficiency of a GPU and an FPGA-based edge platform using a set of CNNs with different computing requirements.