Advancing AI-readiness and Machine-Actionability in the EOSC Ecosystem

Expected Outcome:

Project results are expected to contribute to the following outcomes:

• EOSC will be advancing AI-readiness and Machine-Actionability (MA) in the ecosystem by offering AI-ready federated infrastructure and easy-to-use platform services for EOSC, in order to respond to one of the main challenges of research infrastructures for AI in science, namely the lack of interoperability between AI/ML solutions.
• EOSC will focus on integrating machine-actionable repositories within the ecosystem and demonstrating their reliability and effectiveness by collaborating with repository owners and service providers to implement MA tools and protocols, ensuring seamless integration with the EOSC EU Node.

Scope:

Today, the sustainable FAIRification of data can be a bottleneck towards the goal of a European web of FAIR data and services. The use of AI/ML can significantly help in the process of FAIRification, data curation and data quality assurance, close to the source of the data. EOSC shall promote actions that give incentives to further advance AI-readiness and Machine-Actionability in the EOSC federation for FAIRification and to support their application.

European researchers need access to compute and repository services to develop, train and validate AI/ML models, in line with the GenAI4EU initiative and other key initiatives, like the Apply AI strategy, based on AI-ready research data from research infrastructures and third-party repositories. The proposed infrastructures should complement the EOSC EU Node capacity and be able to scale to a large number of users within transnational access to high-value datasets from national and European Research Infrastructure and e-Infrastructure ecosystems.

AI-based assistance tools shall be customised and trained for the discovery and composition of open science resources into custom workflows allowing researchers to discover and interact with open science infrastructures, combining relevant data, software and application assets.

Open Data and Open Research Software are essential for reliable, trustworthy, and transparent AI/ML. They ensure that datasets and algorithms are well documented, accessible, and reproducible, enabling others to validate and understand AI/ML algorithms. This transparency fosters trust, supports ethical standards, and ensures compliance with regulations, particularly important in the field of AI/ML.

The proposals should focus on all following aspects:

• Develop and prototype tools to drive machine-actionability in repositories, data, and services, establishing a network of trusted repositories linked to the EOSC EU Node;
  • formulate open protocols and policies to facilitate effortless data access, transfer, processing, and provenance updates within EOSC's repository and service network;
  • deliver AI-based capabilities to make interoperable AI/ML solutions and facilitate the setup of custom workflows for research data processing;
  • offer tools/services for automatic quality measures of inputs and outputs of the AI/ML models.
• Provide federated infrastructure services for serving AI models integrating horizontal and thematic EOSC nodes:
  • ensure capacities for AI model retraining and inference;
  • take into account the whole research data life cycle, including raw data retention before AI modelling.
• Provide access to an easy-to-use technology platform offering reference implementations and recipes to quickly get started working with AI/ML with limited engineering overhead:
  • promote and apply state-of-the-art AI/ML operational best practices;
  • validate reference implementations and share commonly used recipes within EOSC.
• Establish and/or provide access to existing AI/ML model repositories and services to serve models for retraining of generic models for specific needs for future predictions and reproducibility:
  • create AI/ML model repository and enhance FAIRness of existing AI/ML models;
  • offer services for utilisation of these models, including fine-tuning and inference, thus providing the foundational building blocks for the development of AI applications in EOSC.
• Establish an EOSC AI/ML competency centre for the pooling of expertise and coordinated support on AI/ML use of data, compute infrastructure and AI/ML models for the upskilling and technical support of EOSC users and research operators, as a strategic asset that will enable a new paradigm for science production.

The proposers should take into account and leverage on the results of relevant projects in the field, including EOSC Data Commons^[1], and the other initiatives and projects contributing to the development of EOSC, especially in the area of machine-actionability and data FAIRification.

This topic implements the co-programmed European Partnership for the European Open Science Cloud.

[1] EOSC Data Commons, Grant agreement ID: 101188179 https://cordis.europa.eu/project/id/101188179