The increasing prevalence of connected devices, from smart home appliances to industrial sensors, is driving a shift in artificial intelligence processing, moving it away from centralized cloud computing and towards “edge computing.” This evolution, particularly the integration of AI into edge networks – known as Edge AI – is gaining momentum in Europe, offering faster response times, enhanced data privacy, and reduced reliance on constant cloud connectivity.
Historically, processing the vast amounts of data generated by the Internet of Things (IoT) required powerful cloud-based platforms like Amazon Web Services and Google Cloud Platform. These platforms host computationally intensive AI models, including recently developed Foundation Models (FMs) – machine learning models trained on broad datasets adaptable to various tasks. However, transmitting data to the cloud introduces latency, potentially delaying responses by hundreds of milliseconds or even seconds, depending on network conditions. This delay can be critical in applications requiring real-time decision-making.
Edge computing addresses this limitation by bringing computational resources closer to the data source – within a building, on local gateways, or at nearby micro data centers. While edge devices have less processing power and storage than cloud servers, advancements in techniques like Split Computing are enabling the deployment of complex AI models across distributed edge nodes. Split Computing partitions deep learning models, allowing different parts to run on different devices, even across geographically dispersed locations.
The benefits extend beyond speed. Processing data locally enhances privacy, particularly for sensitive information. Federated Learning, a key Edge AI technique, allows machine learning models to be trained directly on edge devices without sharing raw data, transmitting only model updates to a central server for aggregation. Once trained, these models can perform inference – generating insights from data – locally, further protecting data privacy. What we have is particularly valuable for industries and small to medium-sized enterprises (SMEs) seeking to leverage Large Language Models (LLMs) while maintaining control over confidential data, such as operational status and maintenance predictions for industrial machinery.
Despite the advantages, deploying Edge AI presents challenges. Unlike mature cloud platforms, a standardized infrastructure for large-scale edge application deployment is still emerging. Telecom providers are beginning to leverage existing infrastructure at antenna sites to offer compute capabilities, but managing these distributed resources – often comprised of numerous low-capacity servers and devices – is complex. Maintenance complexity is a significant barrier to wider adoption.
Researchers are actively working to address these challenges. The Horizon Europe project PANDORA, in collaboration with partners across Europe, is developing an AI-driven framework that provides AI models and computing resources as a service across the IoT-Edge-Cloud continuum. This framework dynamically allocates workloads based on performance metrics like energy efficiency, latency, and computational capacity, optimizing resource utilization and ensuring continuous operation of AIoT applications. The system selects models based on specific intent requirements, such as minimizing energy consumption or reducing inference time, and continuously monitors and updates them to maintain optimal performance.
The development of PANDORA and similar initiatives signals a growing effort to unlock the potential of AIoT systems in smart spaces, including homes, offices, industries, and hospitals, by intelligently allocating resources across the IoT-Edge-Cloud spectrum.
