Bringing AI to the edge and “building intelligence everywhere”

Bringing AI capabilities to the edge of the network can enable high-performance computing capabilities such as low latency, security, resilience, and bandwidth to the area where data is being produced and exchanged and IoT devices are located.

According to Jovanka Adzic, Innovation Lead at TIM and Editor of AI@EDGE Overall Architecture, edge computing is crucial in enhancing end-user devices in use cases such as cloud robotics, Industry 4.0, smart city, smart mobility, and extended reality. It will allow for an increase in system response times and better performance. One of the main challenges to achieving this is the orchestration of the complex and distributed cloud edge infrastructure, said Adzic.

The EU-funded AI@EDGE innovation three-year project started in January 2021, bringing together 20 companies from across the industry, including academia and SMEs. The project focused on AI in beyond 5G networks and how AI/ML frameworks could deliver closed-loop network automation while preserving privacy in multi-stakeholder environments. The project looked at creating a distributed and decentralized connect-compute platform to support a diverse range of AI-enabled network and end-user applications.

A key objective was to implement reusable data and ML pipelines in the AI@EDGE system architecture to avoid AI silos. The AI@EDGE architecture includes AI “on platform” and AI “in platform”, data and ML pipelines and governance, end-to-end system orchestration and management, and distributed connect-compute platform.

Four key use cases were highlighted including, self-driving cars, beyond-visual-line-of-sight drones, in-flight entertainment services, and secure orchestration of Industry 4.0 networks.

The project defined AI Functions (AIF) with specified descriptions to be used by a multi-tier orchestrator to deploy and orchestrate AIFs across distributed connect-compute platform. The system architecture includes some “in platform” AIFs that enable intelligent placement of services at the edge and anomalous event detection.

“The AI native architecture, similar to that of a cloud-native architecture, will provide an environment with intelligence everywhere, distributed data infrastructure, and exposure of AI services that is all managed with zero-touch automation,” Adzic said.

Adzic closed the presentation by highlighting a number of challenges that need to be addressed. This included work on energy performance, integration of both device and edge intelligence, software development for an intelligent edge, and near real-time data analytics processing.