Enabling an edge-computing architecture with AI is seen as a way forward for advances in strategic applications. And at the advent of 5G network speeds, AI is seen as essential to the endpoints. A new network paradigm based on virtualization enabled by Software Defined Networking (SDN) and Network Function Virtualization (NFV), presents an opportunity to push AI processing out to the edge in a distributed architecture, suggests a recent report from Strategy Analytics. Three types of edge computing are foreseen: device as the edge, in which an IoT device generates and consumes data and has embedded AI that can send and receive data to and from additional AI systems; enterprise premise network edge, that can support AI processing on a piece of hardware in a vehicle, drone or machinery, and can collect and process data from smart devices; and operator network edge, with an AI stack/platform to host applications and services, which may be located at a micro data center in a radio tower, edge router, base station or internet gateway. "One of the challenges this new networking paradigm creates stems from the fact that the edge of the network is consistently shifting and moving," stated Caroline Chan, VP and GM, 5G Infrastructure Division, Network Platform Group, Intel, sponsor of the report. In a cloud/client architecture, the link between the centralized cloud and the client has become a bottleneck as more data is processed and network latency increases, causing too much of a time delay.

AI-powered visual inspection pioneer Neurala today announced that its vision AI platform, Brain Builder, has been optimized to enable edge learning for manufacturing and other visual inspection use cases. The latest Brain Builder SDK will debut as a hackathon partner of Bosch ConnectedExperience (BCX), Europe's largest IoT hackathon, in Berlin, Germany from February 17-19. Brain Builder is the only edge learning-enabled AI tool that allows users to build, deploy and analyze custom vision AI solutions with instant feedback on performance and continuous learning in deployment. "Traditional approaches to training deep neural networks (DNN) often fall short in deployment when the network encounters a new situation at the edge that it was not trained to classify," said Massimiliano Versace, CEO and co-founder of Neurala. "This issue is compounded by the fact that oftentimes, DNNs are trained on a server, and then deployed on a smaller, less powerful compute edge that is incapable of learning new information."