IoT-NGIN has been gracefully passing through its final semester. In the past months, the project has successfully delivered 2 releases of components enhancing IoT technology, an integrated version of the IoT-NGIN framework, as well as intermediate validation results in the Living Labs.
Specifically, the IoT-NGIN project has validated its developments for Enhancing IoT Underlying Technology. Improved 5G coverage using D2D connections has been validated, verifying the possibility of extending the 5G coverage through 5G devices. Deterministic and time-sensitive communication for 5G has been validated through time synchronization for a private 5G network at CMC laboratory as well as in ABB living labs. Moreover, the use of the simplified 5G resources management API has been demonstrated with real backends like a Kubernetes instance, or by running it against simulated backends. Also, the usability as well as the performance of the secure edge cloud framework has been demonstrated by running ML models from the project with the framework’s technologies.
IoT-NGIN also offers a set of frameworks and services aimed for enhancing the intelligence of IoT applications and devices, namely: I) Machine Learning as a Service (MLaaS), as the main IoT-NGIN MLOps platform for IoT, and II) the Privacy Preserving Federated Learning (PPFL) platform, specialized for Federated Learning, and iii) complementary services and frameworks, including the online learning service and the model sharing framework, that are delivered with the MLaaS platform. Extended and new features for online learning have been developed, including:
i) a new internal pipeline implementation based on KServe that provides greater flexibility of reuse in multiple inference scenarios,
ii) the support for eXplainable AI (XAI) to explain how models learn from features,
iii) a learning monitoring system for detecting data drift.
Moreover, our latest work includes a Reinforcement Learning (RL) based implementation for system optimization, deployed within MLaaS, and tailored to optimize electric grids for the Smart Energy LL, as well as a common entry point API that harmonizes the access of third parties to the PPFL platform, common for all the FL frameworks supported. A Model Sharing framework implementation has been also released, that enables an integrity-guarantee batch training of ML models, and their registration within the MLaaS model storage for further sharing and reuse, as well as their conversion into ONNX intermediate model for inference in any environment compatible with ONNX runtime.
IoT-NGIN has also delivered contributions related to enhancing IoT tactile and contextual sensing/actuating through innovative features and functionality for the next-generation Ambient Intelligence IoT. The contributions include diverse methods for device discovery, based on both visual and non-visual methods, including ML-based device recognition through computer vision, Visual Light Positioning and UWB-based localization techniques or a combination of those. Moreover, solutions for convergence to a Digital Twin solution, supporting the monitoring part and pervasive access control have been finalized, ensuring that collected data are usable by the Digital Twin, which means that they represent the necessary asset properties and are of sufficient quantity and freshness in order to create credible context, and that security is ensured at the connection points with both the devices and querying entities. In addition, we have created a Gitlab repository including a set of open-access resources that allow rich interaction with IoT sensors and real-world objects via AR applications. These include both own developments from the IoT-NGIN consortium as well as collected third-party resources (e.g., GitHub projects, tools, online courses) that can become useful for the interested audience in these topics.
Cybersecurity, privacy preservation and trust improvement in the domain of IoT systems have been also approached by IoT-NGIN through a set of contributions. The GAN-based dataset generator has been released for the creation of datasets that assist in addressing attacks against IoT and Federated Learning systems, while the Malicious Attack Detector (MAD) facilitates the detection of cyber-attacks against Federated Learning systems, either at the network level or during the learning process. The IoT Vulnerability Crawler (IVC) monitors IoT nodes and detects vulnerabilities that may be identified in networked devices. Moreover, the Moving Target Defense (MTD) Honeypot Framework deploys the honeypots dynamically, opening detected vulnerabilities in DMZ in order to allow the monitoring of lured attackers’ behaviour. In addition, towards facilitating access to data while ensuring data privacy and trust, Semantic Twins enable semantic descriptions of Digital Twins and the related real-world entities, the Decentralised Interledger Bridge (DIB) enables transactions across different distributed ledgers (DLTs) and the privacy-preserving Verifiable Credential based decentralised on-device access control solution for constrained IoT Devices, together with a QR (Quick Response) code and GS1 (Global Standards 1) Digital Link based discovery mechanisms.
The project has progressed on the Living Lab validation side with intermediate results of the test and validation processes for the 10 Use Cases of IoT-NGIN across Smart City, Smart Agriculture, Smart Industry and Smart Energy. The intermediate results include functional testing of the IoT-NGIN components in the Living Labs in the context of the defined scenarios. So far, the results verify proper and smooth operation of IoT-NGIN in the Living Labs, and the next step is towards evaluating the value brought by IoT-NGIN innovations in the piloted use cases.
In addition, the IoT-NGIN validation potential has been further extended with the admission of 10 new projects under the 2nd Open Call, which ran from July 2022 till the end of September 2022. The new projects follow a “DESIGN-EXPERIMENT-GROWTH” stages programme, while the “design” phase has been already completed, through which 6 projects qualified for the “experiment” phase.
Do not miss our latest project developments available as open source at the IoT-NGIN public group on GitLab.
Stay up to date with our project news for the future technological releases!