Efficient trustworthy AI - making the best of data (AI, Data and Robotics Partnership) (RIA) -

Expected Outcome

Projects are expected to contribute to the following outcomes:

• Optimized AI solutions: optimizing model design and data usage to maximize accuracy and robustness.
• Ensure in general, the pipeline of high-quality, representative, unbiased and compliant training data for AI development in all relevant sectors
• Support data preparation and AI training processes that lead to efficient and more trustworthy AI

Scope

There is a need for AI methods that optimize training and reduce the amount data, the intensity of processing and the operations necessary for training high-quality, trustworthy AI systems. As a consequence, the energy consumption and the environment footprint will also be reduced. Such solutions are of relevance also in the context of embedded and embodied AI, i.e. AI capabilities in robotics and connected devices/objects/embedded processors, including small (down to micro/nanoscale) objects with long-term autonomy.

Proposals should address novel AI methods and training data provision processes, aiming at high quality and reliable AI while minimizing the data needs and manipulations, targeting smart and dynamic end-to-end automation of AI training in the cloud-edge computing continuum, where AI training, AI deployment and data collection/preparation happens at the most appropriate level of the cloud-edge continuum. This will lead to better quality of AI by smart data selection/harvesting/preparation and reduces the need to collect, store, process and transfer large amounts of data and/or large AI models, while reducing energy consumption.

Proposals should address at least one of the following focus areas:

• automated and AI-based mining, harvesting, selection, cleaning, annotation, and/or enrichment/augmentation of data for AI; generating and using synthetic data to reduce the need for large volumes of real and potentially sensitive data; validating the efficiency of these processes in AI systems;
• lighter, less data-intensive and less energy-consuming AI models, optimized learning processes that require less input (data efficient AI) without degrading the quality of the output; machine learning methods and architectures that deal with lower volumes such as transfer learning; one-shot learning; continuous and/or lifelong learning.

Proposals should clearly mention which of the two areas will be their main focus area.

The work should contribute to increasing data efficiency and energy efficiency of AI, and rationalize the provision of data for AI. The work should support appropriate AI paradigms (central, distributed, dynamic, hybrid), responding and adapting easily to the needs of the use situation, and to the changing characteristics, availability and use conditions for data.

Target AI systems should be appropriately evaluated, and results analysed and fed back to ensure continuous improvement of the “data for AI” pipeline.

Multidisciplinary research activities should address all of the following:

• Proposals should involve appropriate expertise in all the relevant disciplines, such as e.g. engineering, data science, computer sciences, mathematics, and where applicable in Social Sciences and Humanities (SSH) and gender expertise.
• Projects should build on or seek collaboration with existing projects and develop synergies with other relevant European, national or regional initiatives, funding programmes and platforms, especially the actions funded in the Digital Europe programme, under the chapter “Cloud, data and artificial intelligence”.
• Contribute to making AI, data and robotics solutions meet the requirements of trustworthy AI, based on accuracy, robustness, safety, ethical principles and reliability, in line with the European Approach to AI. Ethics principles needs to be adopted from early stages of development and design.
• Proposals are expected to dedicate tasks and resources to collaborate with and provide input to the open innovation challenge under HORIZON-CL4-2023-HUMAN-01-04 addressing optimisation. Research teams involved in the proposals are expected to participate in the respective Innovation Challenges.

All proposals are expected to embed mechanisms to assess and demonstrate progress (with qualitative and quantitative KPIs, benchmarking and progress monitoring, as well as illustrative application use-cases demonstrating concrete potential added value), and share communicable results with the European R&D community, through the AI-on-demand platform, Digital Industrial Platform for Robotics and Common European data spaces, and if necessary other relevant digital resource platforms in order to enhance the European AI, Data and Robotics ecosystem through the sharing of results and best practice.

The proposal should describe the characteristics and availability of the data to be used within the project and explain how the possible privacy and IPR issues related to the data are addressed. The provenance, associated metadata and any other contextual information should be collected and maintained to the extent necessary in order to enable validation and support explainable AI and to ensure continuous compliance with applicable legislation (e.g. GDPR, AI act, data act).

In order to achieve the expected outcomes, international cooperation is encouraged, in particular with Canada and India.

This topic implements the co-programmed European Partnership on AI, data and robotics.

Activities are expected to start at TRL 2-3 and achieve TRL 4-5 by the end of the project – see General Annex B.

News flashes