MANOLO – Trustworthy Efficient AI for Cloud-Edge Computing

Toimenpiteet

In the MANOLO project, we:

• Map what end-users, technology providers and AI ethics experts need from trustworthy, efficient AI. We gather this through interviews, workshops, and an online survey, then turn it into MANOLO requirements and system design.
• Develop hardware-aware AI algorithms that combine pruning, quantisation, knowledge distillation, meta-learning and neural architecture search. The goal is models at least three times smaller that cut training energy by more than 30% on at least four hardware platforms.
• Build a trustworthy data-management framework that records where data, models, and algorithms come from and how they change. It detects biased or malicious data and creates compact, high-quality training datasets.
• Optimise AI training, inference, and data across cloud and edge to reduce energy use.
• Benchmark models on at least seven standard and neuromorphic hardware platforms in more than 150 experiments, measuring energy, accuracy, speed, explainability and robustness.
• Embed trustworthiness and legal compliance from the start. We use the Z-Inspection® process, the EU Ethics Guidelines for Trustworthy AI and the AI Act, Data Act and GDPR, with guidance from an Ethics Advisory Board.
• Pilot the MANOLO toolset in four use cases across healthcare, manufacturing and telecommunications, on robots, wearables and smartphones. We stress-test it in the lab and then in realistic conditions.
• Evaluate the use-case results against the original requirements, covering technical aspects (accuracy, energy, data handling, hardware performance) and non-technical ones (ethics, safety, trustworthiness).
• Train developers, researchers, students, and other stakeholders in human oversight and trustworthy AI through workshops, a summer school, and hands-on sessions.
• Publish open-source tools, benchmarks, public deliverables and peer-reviewed articles through Zenodo, GitHub and the AI-on-Demand platform.
• Connect with other Horizon Europe and Digital Europe projects, ADRA networks and EU platforms to share knowledge and keep the results in use after the project ends.
• Advise policymakers with evidence-based best practices for designing and deploying trustworthy, efficient AI, and with recommendations on how ethics and human rights can be built into policy.