DECENTER provides platforms and tools to create AI-based cloud-native applications and operate them in the cloud-to-edge continuum. Cloud-native applications are composed by set of independent modules (containers) that can communicate among them, gather information from sensors, provide inputs to actuator and other active devices or modules. Some of these containers may contain trained AI models, some other may provide other functionalities, e.g. a database, a web-interface, etc.
The DECENTER platform offers the capability to assemble different containers, and to dinamically create an AI-based cloud-native application that is then deployed in the infrastructure. Resource optimisation algorithms can be easily introduced, according to the need of the operator: the consortium has researched a set of solutions that minimise the utilisation of resources in the cloud-to-edge continuum while catering to the specific requirements of each application. It may also be the case in which the resources of the infrastructure provider are not sufficient to deploy applications; for such cases, DECENTER proposed a blockchain-based brokerage platform to host negotiation and seal contracts among parties, leading to automated federation of resources. A monitoring system supports these operations, providing data through which the quality of service that DECENTER can offer is assessed. IoT middlewares integrated within the DECENTER platform offer the capabilities of collecting data from sensors and other appliances directly at the edge of the infrastructure, offering the possibility to analyse them in a timely-effective manner in the fog nodes in which the AI containers have been deployed.
Tools and services are useful to create containers, enable cross-border data sharing, provide security, and much more. The project investigated methods to optimise AI models in order to facilitate their deployment in resource-constrained edge nodes. Concerns related to the management of data were addressed, especially in the context of cross-domain activities, in which the inherent trust offered by blockchains can help. AI applications can benefit from other essential services: DECENTER designed a data model for digital twins and proposed to reuse the data brokerage capabilities of IoT platforms to enhance such digital representation with both sensor data and AI elaborations. Other DECENTER services offer basic tools to create containerised AI applications and to secure the edge nodes and AI applications.
DECENTER Consortium has demonstrated that through its technology it is possible to detect and notify dangers for pedestrian in smart crossing in less than 100ms, reduce of a factor of almost two the CPU consumption of logistic robots running AI methods (thus increasing their availability and reducing energy consumption), increase privacy while preserving safety in smart construction sites, and reduce the size of complex AI models (e.g. less than half the parameters) to run them on resource-constrained edge devices, in order to speed up computation time and save bandwidth.
Outcomes have been disseminated through scientific publications (15 journal and 19 conference papers), and communicated through the project's website and social networks, as well as through international events. Three assets reached a TRL of 7, leading to relatively short-term commercial perspectives. These exploitations take different form, including the creation of a dedicated startup company. Related to project results, 15 patents were also filed. Finally, EU and KR efforts were dedicated to standardisation, with contributions to ISO/IEC standards.
