Sébastien Taylor, Vision Technology Architect at Au-Zone Technologies, presents the “Harnessing the Edge and the Cloud Together for Visual AI” tutorial at the May 2018 Embedded Vision Summit.
Embedded developers are increasingly comfortable deploying trained neural networks as static elements in edge devices, as well as using cloud-based vision services to implement visual intelligence remotely. In this presentation, Taylor explores the benefits of combining edge and cloud computing to bring added capability and flexibility to edge devices.
For example, an edge device can use a locally implemented neural network to address common cases, and utilize larger models in the cloud for unfamiliar events, localization differences and corner cases. Cloud resources can also be used to provide updated neural network models to edge devices.
Taylor explores a cascaded machine learning architecture that takes advantage of both edge and cloud computing to create a system that can dynamically adapt to new conditions. Using image classification use cases, he describes the solution in detail, including system capabilities, applications and design trade-offs.