Title: Neural model compression for edge computing
Authors: Vahid Partovi Nia - Ecole Polytechnique de Montreal / Huawei Noah's Ark Lab (Canada) [presenting]
Abstract: Deep neural networks is an effective tool for many supervised learning tasks, such as voice recognition, object detection, image classification, etc. In 5G technology many optimization tasks rely on effective user behaviour prediction, in which neural networks play a central role. Neural networks tend to have many parameters, in order of millions, or even billions. Therefore, their deployment on edge devices such as cell phones, smart watches, IoT devices, and wireless base stations is a major challenge. We introduce combination of different strategies to simplify neural networks in order to save memory, computation power, and energy.