Leveraging Benchmarking Data for Automated Machine Learning and Optimization

Información del proyecto

AutoLearn-SI

Identificador del acuerdo de subvención: 101187010

DOI

10.3030/101187010

Fecha de la firma de la CE 5 Agosto 2024

Fecha de inicio 1 Marzo 2025

Fecha de finalización 28 Febrero 2030

Financiado con arreglo a

• Widening participation and spreading excellence

Coste total

Sin datos

Aportación de la UE

€ 2 485 960,00

Coordinado por

INSTITUT JOZEF STEFAN
Slovenia

Periodic Reporting for period 1 - AutoLearn-SI (Leveraging Benchmarking Data for Automated Machine Learning and Optimization)

Período documentado: 2025-03-01 hasta 2026-05-31

Resumen del contexto y de los objetivos generales del proyecto

Artificial intelligence has rapidly transformed research and industry across Europe, with the European Research Executive Agency managing over 1,000 AI-related projects totalling €1.7 billion in funding. Automated Machine Learning (AutoML) and Automated Optimization (AutoOPT) — collectively AutoAI — have emerged as critical enablers for democratized AI use, reducing the need for deep expertise and making AI accessible to small businesses and non-specialist practitioners.

Despite this global momentum, a significant gap exists in Slovenia and the wider Central and Eastern European region, where dedicated AutoAI research groups, academic programmes, and training opportunities are largely absent. AutoLearn-SI, led by the Jožef Stefan Institute (JSI) and submitted to the Horizon Europe ERA Chairs call, seeks to address this by establishing a permanent, internationally excellent AutoAI research group. The ERA Chair is Asst. Prof. Dr. Eva Tuba and the coordinator is Asst. Prof. Tome Eftimov.

The project pursues five integrated objectives: establishing the region's first dedicated AutoAI research center at JSI; generating new scientific knowledge across experimental databases, representation learning, and algorithm selection; training researchers in both technical and transferable skills; embedding AutoML and AutoOPT courses into master's and doctoral programmes; and building long-term sustainability through industrial partnerships and follow-on funding.

Expected outputs include 24 peer-reviewed publications, two FAIR benchmarking databases, and ten open-source Python implementations. Industry guidelines aligned with the EU AI Act will support transparent AI adoption, while a focus on algorithm efficiency will reduce energy consumption and promote greener AI practices. Over the longer term, AutoLearn-SI aims to become a self-standing JSI department and a regional hub for AutoAI research and innovation.

Trabajo realizado desde el comienzo del proyecto hasta el final del período abarcado por el informe y los principales resultados hasta la fecha

During the initial phase of AutoLearn-SI (WP1 and WP2, covering months 1–15), the project successfully established the ERA Chair within JSI's Computer Systems Department, putting in place all key governance and management structures, including the Executive Science Committee, Management Board, and Advisory Board. The ERA Chair holder and PhD candidates were recruited in line with the European Code of Conduct for the Recruitment of Researchers. Foundational project documents were delivered, including the Data Management Plan, the Dissemination and Exploitation Plan, and the ERA Chair's strategic sustainability plan. The project's public-facing presence was launched through a dedicated website, social media accounts, and a suite of promotional materials, complemented by press releases and editorial articles targeting the international AutoAI research community. On the scientific side, the three PhD students commenced work on their respective topics — covering experimental databases and meta-features for single-objective optimization, experimental databases and meta-features for multi-label classification, and automated algorithm selection and configuration — producing initial state-of-the-art knowledge reports as the foundation for future AutoML and AutoOPT research. Educational activities were also completed, including a two-day exploratory workshop bringing together leading EU researchers from associated partner institutions, a Science Communication training event, and an AI and Data Privacy & Security training event, all open to national and international students and young entrepreneurs.

Avances que van más allá del estado de la técnica e impacto potencial esperado (incluida la repercusión socioeconómica y las implicaciones sociales más amplias del proyecto hasta la fecha)

AutoLearn-SI has advanced the state of the art in Automated Machine Learning (AutoML) and Automated Optimisation (AutoOPT) through systematic cataloguing, structured knowledge building, and the establishment of shared scientific infrastructure. In single-objective optimisation, the project reviewed and structured existing landscape meta-feature extraction approaches, mapping key gaps including benchmark bias, generalisation limitations, and interpretability deficits. In multi-label classification, a systematic overview of meta-learning frameworks was produced, significantly extending the MLC Dataset Catalogue to 89 datasets and 26 benchmark methods as an open community resource. In algorithm selection and configuration, trajectory-based representation approaches were catalogued, identifying the lack of dynamic run-time representations and limited cross-domain transfer as critical open challenges. Underpinning all three areas is the OPTION knowledge base — a FAIR-compliant, RDF-structured repository of over 226 optimisation algorithms serving as shared semantic infrastructure for the AutoAI community. These activities have supported 15 peer-reviewed publications across 6 international research collaborations.
Looking ahead, Phase II will focus on completing two FAIR experimental databases for SOO and MLC, delivering open-source Python libraries, and consolidating the meta-feature catalogue as a shared community resource. Industry-facing AutoAI guidelines aligned with EU AI Act requirements will be developed, targeting adoption through the Slovenian AI Factory and AutoML 2026 in Ljubljana. Sustainability is secured by over €1,357,676 in confirmed external funding — including MSCA Postdoctoral Fellowships and MSCA COFUND SQUASH — ensuring continuity of the AutoAI research group at JSI beyond the ERA Chair grant. Collaboration with Slovenia's Ministry of Digital Transformation, IRCAI UNESCO, and two national SRIP networks further positions project outputs to contribute to national AI governance and broader EU standardisation efforts.