Holistic Edge AI: From ML Models to Hardware-Aware Deployment

The rapid expansion of Edge Artificial Intelligence (Edge AI) within Industry 4.0 demands engineers capable of efficiently deploying machine learning (ML) models on embedded platforms operating under strict performance and energy constraints. Yet, most engineering curricula continue to separate algorithmic instruction from hardware-aware design, resulting in a persistent skills gap for real-time, resource-limited applications.To address this challenge, From Algorithm to Hardware: Machine Learning in Embedded Systems was developed as a two-week intensive program combining ML optimization techniques, such as pruning, quantization, and knowledge distillation, with hardware-oriented topics including power and signal integrity and electromagnetic compatibility. The course bridges theory and practice through a multidisciplinary, hands-on approach in which participants complete a full FPGA-based workflow from algorithm design to on-board deployment, fostering system-level thinking and early design trade-offs.Quantitative results from participant assessment showed substantial learning gains, with average self-assessed expertise increasing from 1.9–2.5 before the course to 4.3–4.7 after training across six technical domains and formative quizzes. Qualitative reflections highlighted the value of the end-to-end workflow, peer discussion, and exposure to real-world design constraints in strengthening interdisciplinary competence.Overall, the proposed framework demonstrates how practice-based, cross-domain learning can enhance engineering education by uniting algorithmic and hardware perspectives. It provides a replicable model for developing Edge AI expertise in Industry 4.0 contexts while supporting the transition toward Industry 5.0, where human–AI collaboration, ethical awareness, and sustainable design are central.