Artificial intelligence (AI) and the Internet of Things (IoT) are transforming industries and society by enabling fast and smart decision-making, with little or no human intervention. As the adoption of these technologies grows at an ever quickening pace, we are also observing an increasing resort to cloud computing solutions for offloading the computational burden required to process large amounts of data produced by IoT devices, and by advanced AI models developed to deal with that data.

Unfortunately, cloud computing services, usually available in remote third-party infrastructures, often fall short in meeting the requirements of latency- and privacy-sensitive applications. Recently, the fog computing paradigm has been proposed as an effective means to bridge the gap between the cloud and IoT technology domains, making computing resources more easily available closer to where data is produced, hence reducing latency and avoiding the exchange of sensitive data with third-party entities.

Due to the relevance of these topics, Atos decided to get on board in DECENTER project (www.). This research project provides a fog computing platform to orchestrate cloud-to-edge resources that provide all the necessary tools to create and operate AI-based workloads close to the IoT infrastructure, i.e., where the data is produced. With respect to cloud-based solutions, DECENTER enables real-time data analytics and low-latency actuations, while also ensuring privacy by design.

This platform is optimized for hybrid decentralized AI models, coping with data coming from different sources. It focuses on four innovation goals, aiming to advance the fog paradigm to its next evolutionary step from four different aspects that are complementary to each other:

  1. Develop a robust fog infrastructure for the deployment of AI applications;
  2. Ensure cross-border federation and interconnectivity of edge resources using blockchain technologies;
  3. Enrich the IoT ecosystem to process large quantities of heterogeneous data at the edge of the infrastructure;
  4. Develop decentralized AI models that exploit the full potential provided by fog computing.

Atos is contributing to the project in the detection of anomalies in edge devices (L-ADS component) and service level agreements between the (SLA Lite component).

DECENTER is being validated in four very relevant use cases where the platform benefits can be seen in a tangible way:

  1. Smart city crossing safety: Monitoring road crossings to spot and signal in real-time potential dangers that might put pedestrian safety at risk;
  2. Robotic logistics: Novel, cost-effective, robotic indoor transport solution especially suited for warehouses relying on time-sensitive edge-based automation;
  3. Smart and safe construction: AI methods for video stream analysis in a construction site scenario to provide early warning against accidents and more;
  4. Ambient intelligence: member verification service at the edge using AI models without sending any personal information to the cloud for IoT-based services in devices.

The DECENTER high-level architecture (see figure) defines all the necessary tools to create and operate AI-based workloads in a heterogeneous, distributed and opportunistically created fog computing infrastructure, covering the whole cloud-to-edge continuum. The DECENTER architecture includes platforms and services to support the entire cycle of creation and operation of AI applications. Services (in red) are introduced to support the creation of AI-based cloud-native applications starting from models that can be easily retrieved, managed and shared. The platforms (in blue) are then used to capture data from sensors and devices to orchestrate the fog computing resources according to the needs of the AI applications, to monitor their behaviour and to acquire and manage resources from third parties which can be necessary to cater to the requirements of the applications.