dAIEDGE attracts 146 researchers from across 21 countries
Thanks to activities such as webinars and outreach events, we received 146 applications from researchers worldwide, highlighting the global appeal of our programme. From among these, 36 complete proposals were submitted, representing 21 countries.
Ada Guzey Muhendislik Yazilim Mekatronik ltd.
Hosting Institution: CSEM
Acronym: ENSURE [Edge ai-eNhanced Safe aUtonomous navigation using contRol barriEr functions]
Giacomo Donati
Hosting Institution: UGLAS
Acronym: Unsupervised Continual Learning on the Edge for Condition Monitoring
This research project focuses on the hardware–software co-design of accelerators for on-device adaptation of deep neural networks (DNNs) at the edge. The goal is to enable efficient, low-latency, and energy-aware adaptation directly on resource-constrained devices, avoiding reliance on cloud resources.
Intelligent Systems Hub
Hosting Institution: INRIA
Acronym: DRIVE-AI (Drone Real-time Insights and Vision for Edge AI)
The project introduces DRIVE-AI, a developed prototype design that supports integration of advanced multimodal AI models to run directly on resource-constrained devices - on-board, off-line, and entirely independent of cloud infrastructure. It does so through a structured, edge-first integration and workflow optimisation pipeline, tailored for privacy-sensitive environments with limited connectivity and strict compute budgets.
Kosta Pavlović
Hosting Institution: UNIMORE
Acronym: SEAL
SEAL develops a resource-efficient, AI-driven video watermarking solution optimized for real-time deployment on edge devices. By addressing the computational limits of state-of-the-art deep-learning watermarking, the project enables robust and high-quality data protection under strict hardware constraints. SEAL embeds watermarks directly at the data source, thus preventing manipulations immediately at the point of data generation.
Laura Acosta García
Hosting Institution: VICOMTECH
Acronym: Enhancing Realistic Warehouse Management with RL at the Edge
This solution proposes a learning-based warehouse management approach in which decision-making is formulated as a Markov Decision Process and optimized through reinforcement learning. The solution operates within a configurable simulation environment that captures stochastic item flows, order generation, and heterogeneous layouts, enabling adaptive control under diverse operating conditions.
Mateusz Piechocki
Hosting Institution: UNIMORE
Acronym: Continual learning acceleration with on-device training on resource-constrained hardware
The project aims to create a robust, privacy-preserving AI framework tailored for resource-constrained edge devices within Smart City infrastructure. By utilizing hardware-accelerated continual adaptation strategies and federated learning, the system will allow devices to adapt to environments and weather conditions without relying on centralized cloud computing or compromising user privacy.
Sotirios Athanasoulias
Hosting Institution: FORTH
Acronym: ENERGIZE
ENERGIZE enhances NILM with computationally efficient, edge-optimized AI models, delivering accurate, low-latency appliance-level disaggregation through a decentralized, privacy-preserving solution suitable for scalable real-world deployment.
Umea University
Hosting Institution: SU-CNRS
Acronym: Edge intelligence for the Security of Smart Systems (Ei4SoS)
