Energy-efficient artificial intelligence for sustainable Internet of Things and edge applications
Integrating the analytical and decision-making capabilities of artificial intelligence and the connectivity and data collection capabilities of the Internet of Things (IoT), Artificial Intelligence of Things (AIoT) enables smarter, more responsive devices. By enabling these systems to analyse data and make autonomous decisions in real-time using local computing resources, AIoT enhances everything from smart homes and healthcare to industrial operations and automotive systems. A big challenge, however, is to effectively utilise the vast amounts of data generated in IoT. This sheer data volume requires substantial computational power for processing, and the complexity of the algorithms involved complicates matters. The EU-funded VEDLIoT project addresses these challenges by developing an IoT platform that uses deep learning (neural network) algorithms distributed across various levels of the IoT infrastructure – edge, fog and cloud computing. “Our goal has been to develop innovative energy-efficient deep learning methodologies specifically tailored to distributed AIoT applications – operating across multiple locations or devices. We focused not only on optimising algorithms but also addressed the intrinsic safety and security challenges these systems present,” notes project coordinator Jens Hagemeyer.
Enabling the IoT to learn
A key component of VEDLIoT’s approach has been the development of a cognitive IoT hardware platform, referring to its capability to learn from data and reason to make decisions. Leveraging state-of-the-art microservers with optimised hardware accelerators, the platform was designed to be modular and scalable. This means that it can be customised to meet the diverse requirements of various applications. This flexibility across different use cases helps enhance energy efficiency and system performance. “By enabling customisable hardware configurations, VEDLIoT ensures that IoT and edge devices can operate sustainably. This should help extend their effective lifespans across various levels of the computing infrastructure (computing continuum), rendering them better suited to a broad ecosystem of the modern applications they serve,” states Hagemeyer. The VEDLIoT hardware platform combines specialised hardware, diverse hardware accelerators tailored for specific applications and advanced security techniques using trusted execution environments and distributed attestation. The project also provides a design framework for safety-critical AI systems, ensuring they meet stringent safety requirements. VEDLIoT’s cross-vendor solutions enhance flexibility and adaptability, reducing reliance on proprietary platforms and promoting broader adoption of AIoT technologies across various industries.
Testing technologies in different use cases
VEDLIoT has successfully validated its technologies in various applications, including automotive, industrial IoT and smart home environments. Key demonstrations include a self-sustained smart mirror, pedestrian detection and avoidance, electric arc detection for DC distribution systems and predictive maintenance for electric motors. Ten applications were also developed through an open call, covering areas such as laser welding and honey testing. “We achieved over a tenfold improvement in performance and energy efficiency across all key application areas by optimising hardware components and accelerators as well as toolchains and models,” highlights Hagemeyer. These results highlight VEDLIoT’s contributions to advancing AIoT technologies by delivering scalable, efficient and secure solutions. Project impact is expected to be substantial, setting new benchmarks in performance, energy efficiency and security for AIoT, edge and cloud applications.
Keywords
VEDLIoT, AIoT, deep learning, edge, artificial neural networks, cloud, artificial intelligence of things, hardware accelerator, energy-efficiency
