The era of Edge AI
Deep Learning (DL), a component of Artificial Intelligence which mimics neural connections of a human brain, is moving from Cloud to the Edge in order to decentralize and operate on billions of devices. Edge AI in this context refers to the computation that takes place locally on a device, where the data is generated and analyzed. The diffusion of devices and their increasing power capabilities highlights the strategic importance of the AI at the Edge, with many advantages being offered by this technology. Edge AI guarantees the protection of data by eliminating privacy issues typically associated with network data streaming. Embedded AI applications are also highly responsive with very low latency, and can drastically reduce environmental impacts caused by carbon emissions deriving from Cloud energy consumption. The development of a DL Edge AI application involves two main steps: neural network model creation, followed by its deployment to the Edge. A DL model learns from data, and this training step can be performed with the aid of standard library frameworks. The deployment of DL models at the Edge represents a challenge, however, as it involves large and computationally expensive processes.
After a comparison between distributed and centralized AI, we continue with a presentation on the five steps methodology used to build a DL model and to fit it into an Edge device with limited memory and power processing capabilities.
Event on digital.makerfaire.eu
Giuseppe Messina received his MS degree in Computer Science in 2000 at the University of Catania for a thesis on Statistical Methods for Texture Discrimination. He received his Ph.D. in 2011 in Computer Science at the University of Catania for his research in Advanced Techniques for Image Analysis and Enhancement. Since March 2001, he has been working as a software designer for System Research and Applications in STMicroelectronics. Giuseppe was member of the Image Processing Laboratory at the University of Catania. His research interests include Image Analysis, Image Quality Enhancement, Forensic Imaging and Robotics. His most recent efforts include Embedded Software for drone flight control units and robotics applications. He is author of several papers and patents, and reviewer for several international journals and conferences. He is an IEEE member since 2007 and was nominated Senior Member in 2017.
Ivana graduated in Computer Engineering from the University of Palermo in 2002 and joined the System Research and Application unit for STMicroelectronics in the same year. Since then, she has been involved in different R&D activities. She is working on an AI research project to develop audio speech solutions for embedded platforms involving Keyword Spotting and Vocal Command Recognition. Ivana has authored several patents and papers.