At dAIEDGE, our Meet the Winners series continues with a project focused on making AI more efficient, sustainable and deployable on real hardware. In this edition, we feature Juan José Rodríguez Andina (University of Vigo) and Roberto Fernández (Logicmelt Technologies), winners of the Second Open Call, and their project QUAD.
QUAD explores how to bring deep learning models closer to real-world deployment by reducing their computational and energy requirements. The project focuses on model compression techniques such as quantization, aiming to significantly reduce resource usage while maintaining performance for edge AI applications.
The use case selected for QUAD is vehicle detection, where optimized deep convolutional neural networks are implemented on FPSoC devices using the PYNQ framework. This enables a direct comparison between traditional GPU-based inference and FPGA-based solutions, highlighting the potential of more energy-efficient and cost-effective alternatives for real-world AI deployment.
The project benefits from a strong collaboration between academia and industry. On one side, the team at the University of Vigo brings extensive expertise in FPGA systems and edge deployment; on the other, Logicmelt Technologies contributes experience in AI model training and compression, ensuring a full pipeline from model design to hardware implementation.
During his journey within dAIEDGE, Juan José Rodríguez and Roberto Fernández developed their work in collaboration with Deutsches Forschungszentrum für Künstliche Intelligenz, acting as hosting institution. We would like to sincerely thank DFKI for their support, mentorship and access to expertise in AI and hardware systems.
Throughout the seven-month programme, the team also benefited from continuous guidance from dAIEDGE mentors, who helped them overcome key implementation challenges, particularly in representing and deploying compressed models on FPGA architectures. In addition, access to the dAIEDGE-VLab allowed them to test their solutions on diverse hardware setups, accelerating validation and improving practical readiness.
QUAD exemplifies the mission of dAIEDGE: bridging the gap between research and deployment to enable efficient, scalable and environmentally conscious AI systems.