Exploring the Future of Foundation Models: Key Insights from the TDW on Foundation Models

Exploring the Future of Foundation Models: Key Insights from the TDW on Foundation Models

On 10 July 2025, the ELIAS, ELLIOT, and ELSA projects co-organised a Theme Development Workshop (TDW) focused on Foundation Models in Thessaloniki, Greece. The event was hosted by the Information Technologies Institute of the Centre for Research and Technology Hellas (CERTH). The hybrid-format workshop brought together over 90 in-person participants and more than 120 online attendees, including AI researchers, industry professionals, and students. The four-hour event featured three thematic sessions, each spotlighting cutting-edge research, real-world applications, and critical discussions around the development and deployment of foundation models.

While the global narrative around Foundation Models—powerful, resource-intensive AI systems such as GPT-4 and DALL·E—has largely centred on technology giants, this workshop highlighted Europe’s growing momentum in developing open, community-driven alternatives, with notable advances in training, applications, and security.

A European Perspective on Foundation Models

The event commenced with remarks from Elisa Ricci (University of Trento & Fondazione Bruno Kessler), who outlined the workshop’s objectives: to explore the technical, societal, and application-related aspects of Foundation Models—large-scale AI systems that generalise across tasks and domains. Ricci emphasised the importance of European collaboration, public supercomputing infrastructureopen science, and inclusive design in shaping the future of these models.

European grassroots efforts are thriving. Projects such as LAION and EleutherAI demonstrate that large-scale, open-source datasets and models are not only possible but already successful. As highlighted in the workshop, tools such as OpenCLIP (an open alternative to OpenAI’s CLIP) are widely downloaded and used globally, demonstrating the impact of strong ideas and collaboration.

Session 1: Training Foundation Models

Moderated by Elisa Ricci, this session explored scaling laws, efficiency, and data strategies in training large AI models.

Jenia Jitsev (Jülich Supercomputing Centre & LAION) delivered a keynote on scaling laws and generalisation in open foundation models. He emphasised the importance of reproducible scaling laws to predict performance, compare learning procedures and systematically search for learning with stronger generalisation and transferability, highlighting work on OpenCLIP, Re-LAION, openMaMMUT and OpenThinker. A key message was that academics can go surprisingly far and reach up to so-called hyperscaler closed labs in the industry with access to vast public supercomputing resources, strong ideas, and a collaborative, transparent open-source spirit – a compelling case for providing further increased support to open, academic AI efforts.

Frank Hutter (Prior Labs & ELLIS Institute Tübingen) introduced recent innovations in Tabular Foundation Models, a domain often overlooked in mainstream deep learning. He showcased TabPFN and TabPFN v2, which outperform traditional machine learning approaches such as gradient-boosted trees in sectors like finance and healthcare. Hutter demonstrated how synthetic data generation can enable powerful pretraining, proving that domain-specific foundation models have broad potential beyond NLP and vision.

Cees Snoek (University of Amsterdam) presented NeoBabel, a multilingual foundation model for image generation that natively understands six languages: English, Chinese, Dutch, French, Hindi, and Persian. NeoBabel tackles a major challenge: multilingual multimodal data is scarce. The team enhanced an English-only dataset using LLMs for translation and detailed recaptioning, thereby bootstrapping a multilingual dataset from scratch. The model is fully open-source, offering not only checkpoints but also a curated dataset and an extensible toolkit for reproducible research.

📄 NeoBabel paper: arXiv link

🔗View Slides

Session 2: Ethical & Safe Foundation Models

Chaired by Lorenzo Baraldi (University of Modena and Reggio Emilia), the second session explored the security and societal risks associated with foundation models.

Mario Fritz (CISPA Helmholtz Center for Information Security) delivered a comprehensive talk on the security and safety landscape, covering issues such as prompt injection attacks, trust in code-generation models, and the dynamics of agent-based negotiation and collaboration. He explored how foundation models can amplify both risks and opportunities, stressing the need for transparent alignment strategies and automated red teaming.

Session 3: Applications of Foundation Models

Moderated by Dimosthenis Karatzas (Computer Vision Centre – CVC-CERCA & Autonomous University of Barcelona), this session showcased real-world applications of foundation models in robotics, video understanding, and industrial domains.

 

Marc Pollefeys (Microsoft & ETH Zurich) provided a comprehensive overview of Spatial AI foundation models in 3D environments. His talk ranged from robot manipulation to world models for autonomous driving, highlighting progress on the recent GEM models and their application in real-world robotics and perception.

Matthieu Cord (University of Sorbonne & VALEO) presented innovations in generative video pretraining through the VaVIM–VaVAM models. By reducing token counts and shifting from discrete to continuous tokens, the team achieved significant gains in both training speed and performance, pushing the boundaries of video foundation models for automotive and control systems.

Dario Garcia-Gasulla (Barcelona Supercomputing Centre) concluded the session with an in-depth look at training and evaluating LLMs using European HPC resources. He addressed post-training techniques, emphasising the importance of robust evaluation benchmarks, especially in regulated domains such as healthcare, chip design, and secure code generation. Garcia-Casulla also highlighted the increasing role of European AI Factories and Gigafactories in providing accessible compute and sovereign infrastructure for open AI development.

Building a Collaborative Future

Interactive Q&A segments followed each talk, with discussions centred on different topics. The event concluded with a wrap-up session led by Ricci, Karatzas, and Baraldi, who emphasised the importance of cross-sectoral collaboration to ensure that foundation models are developed in ways that are safe, inclusive, and aligned with European values.

Key takeaways from participants included:

  • Support for open, multilingual, and multimodal models
  • Investment in European computing infrastructure to level the playing field
  • Stronger integration of ethics, regulation, and societal perspectives in AI development
  • The value of synthetic data and innovative training techniques to democratise access

The workshop fostered vibrant discussion and provided valuable networking opportunities, culminating in a light lunch where participants continued to exchange ideas informally.

Watch the event recording here!

Looking Ahead

This TDW marked the second in a series of thematic workshops organised by ELIAS, in collaboration with the ELLIOT and ELSA networks. It sent a clear message that Europe is not merely observing the foundation model revolution—it is actively shaping it.

Despite ongoing challenges in data availability and computational resources, Europe’s commitment to open models, responsible design, and regional relevance is yielding tangible results. Initiatives like Laion, NeoBabel, OpenCLIP, and AI Factories illustrate that a distributed, democratic AI future is not only possible—it is already underway.

This workshop was jointly organised by the ELIAS , ELLIOT & ELSA projects.

European Large Open Multi-Modal Foundation Models For Robust Generalization On Arbitrary Data Streams – ELLIOT (GA No. 101214398 ) aims to enhance general-purpose AI by developing large-scale, open multimodal foundation models with strong spatio-temporal understanding. Led by top European academic and industrial labs from the ELLIS and LAION communities, the project targets underrepresented time-relevant modalities such as industrial time series, remote sensing, and health data. Both real and synthetic data will be used, sourced from consortium partners and European Data Spaces, with synthetic data generated using current and novel generative AI methods. European HPC resources are integrated to support large-scale model training. 

European Lighthouse on Secure and Safe AI – ELSA (GA No. 101070617) is a growing network of excellence that spearheads efforts in foundational safe and secure AI methodology research. ELSA’s founding members include European experts in all aspects of safe and secure AI, with particular focus on technical robustness, privacy preserving techniques and human agency and oversight. In addition, ELSA brings on board research and industry experts in six different sectors that are key application areas of safe and secure AI. ELSA builds on and extends the internationally recognised and excellently positioned ELLIS (European Laboratory for Learning and Intelligent Systems) network of excellence. 

AI Launchpad Batch #2

AI Launchpad Batch #2

Centralised Application Process

Say goodbye to the hassle of applying to multiple incubators. With AI Launchpad, you can apply through our central portal and gain access to a network of premier AI startup ecosystems.

A Network of AI Experts and Entrepreneurs

Leverage a pan-European network of AI experts and entrepreneurs. Access centralised learning from ten hubs, bridging the gap between science and entrepreneurship

Immersion and Expansion

Participate in either the Spring or Fall season by joining a curated 1–2 week visit to a leading European AI hub. The program is tailored to your market goals and maturity, connecting you with investors, corporate partners, and key ecosystem leaders. To maximise impact, you’ll also take part in targeted sessions before and after your visit.

Batch #2

Are you elegible?

Teams of 2-4 entrepreneurial students/researchers.

Completing higher education at a European University.

Aiming to develop AI technologies into marketable products/services.

Goal: Launch a VC-fundable start-up with European/global impact.

Our Programme

Visiting Hub Program

The two-week immersive program designed specifically for AI startups looking to navigate and strategize their entry into the European market. This initiative offers participants a deep dive into the unique challenges and opportunities of the European landscape, facilitating tailored growth and expansion strategies

Market Exploration
Network Building

Understand Market Dynamics

Analyse the current trends and economic conditions in the German and broader European markets.

Engage with Thought Leaders

Participate in discussions with leading thinkers in the AI and tech industries.

Identify Sector Opportunities

Pinpoint sectors with high growth potential and demand for your AI solutions.

Form Strategic Alliances

Explore opportunities for forming alliances with established businesses that can offer complementary strengths.

Assess Competitive Landscape

Evaluate local competition to strategize positioning of your AI startup.

Cultivate Relationships with Local Entrepreneurs

Forge connections with local startups and entrepreneurs for cross-cultural insights and collaboration.

Recognise Regulatory Challenges

Learn about specific regulatory requirements and hurdles in Europe that could impact market entry.

Initiate Investor Dialogues

Meet potential investors to discuss funding possibilities and gain financial insights.

 

The local accelerators

Participants can apply to one of our 10 local hubs to be part of a 1-2 week visiting hub program, allowing them to fully immerse themselves in a vibrant ecosystem. This period is crucial for deep engagement with the community and iterative refinement of innovative projects.

Choose between one of ten local accelerator hubs, each offering a unique and tailored programme for your start-up:

INiTS | Vienna, Austria
DTU Skylab | Lyngby, Denmark
Campus Founders | Heilbronn, Germany
ETH AI Center & Talent Kick | Zurich, Switzerland
KTH Innovation | Stockholm, Sweden
XPLORE Venture Creator | Munich, Germany
HPI | Potsdam, Germany
CVC | Barcelona, Spain
Yes!Delft | Delft, Netherlands
Cyber Valley | Tübingen, Germany

How does it work?

Nomination

Startups are referred by local facilitators. No open calls, just quality.

Onboarding

Join 2–3 online workshops to align goals, prepare assets, and connect early.

Immersion

Spend 2 weeks in a new European hub, engage in curated meetings, events, and VC introductions.

What to Submit

  1. Your Startup Overview — Share a clear and compelling pitch deck or one-pager that highlights your vision, product, and market opportunity.
  2. Team Introduction Video (1 minute) — Introduce the people behind the startup. Keep it simple, genuine, and authentic. Record your video, upload it to Loom or YouTube, and include the link in your application.
  3. Nomination Form — Once you have your materials ready, complete the application form and submit your nomination.

Deadline: August 31, 2025 at 23:59 CET

We can’t wait to discover your story — and learn how your AI venture is shaping the future.

More info under: www.ai-launchpad.eu | For enquiries regarding nominations, programmes, or collaborations, please contact:

AI for Trade Challenge

AI for Trade Challenge

Competition Details

About the Challenge

The AI for Trade Global Challenge invites data scientists, economists, and machine learning experts from around the world to help shape the future of trade prediction. Teams will have to submit a forecast of the trade flows for the United States and China for the month of October of 2025. The forecast should provide trade values in dollars for the top 20 sources and destinations for the top 200 exports and imports of the United States and China (at the HS4 product level).

Eligibility
  • Participants may enter in a team of 1 – 5 persons
  • Teams must register to express their interest to participate and also to get access to the training data provided by the OEC
  • Each participant must belong to a single team
  • All participants must be able to identify themselves as natural persons
  • Participants are required to submit:
    1.  a plain text csv file with their forecast, in the following format: “Country1” , “Country2”, “ProductCode”, “TradeFlow”, “Value”
    2.  a 2-page long method description is also required that focuses on how the model was created and implemented.
Dataset & Resources

Historical trade data for the US and China will be made available for all months of the year 2023 and 2024.

Z
Expected Outcomes

We are looking for the most accurate trade prediction for October 2025.
By joining the challenge, along with gaining valuable professional experience, you’ll also have the chance to win exciting prizes.

 

Key Dates
  • Registration starts on July 23 , 2025
  • Submission deadline: October 31, 23:59 CET
Evaluation Criteria
  • How Entries Will Be Evaluated: Predictions about the trade flows of the US and China will be evaluated respectively against trade data published by each of these countries (the US and China).
    The winning team is the one with the best forecast according to sMAPE (symmetric mean absolute percentage error).
    To be eligible for winning, all forecasts must pass a minimum requirement of having a higher accuracy than a forecast based on using raw trade data for July 2025 to predict the flows for October 2025 (historical data shows these baseline sMAPEs to be about 35% to 45%).

  • Judging Panel: TBD

Prizes

First Prize:
3000 USD
Free OEC Premium accounts for 1 year for all team members.

Second Prize:
2000 USD
Free OEC Premium accounts for 1 year for all team members.

Third Prize:
1000 USD
Free OEC Premium accounts for 1 year for all team members.

Have Questions?

For further information or queries, please contact:
📧 ai4trade@centerforcollectivelearning.org

Partners and Support
  • The Observatory of Economic Complexity
  • Poverty and Equity, the World Bank
  • Trade Practice, the World Bank
  • Asian Development Bank
  • Fundación Cotec, Spain
    Academic Partners
    • European Lighthouse of AI for Sustainability
    • Complexity Economic Group, University of Oxford
    • Corvinus Institute for Advanced Studies, Corvinus University of Budapest
    • Institute for Advanced Study in Toulouse (IAST), Toulouse School of Economics​
    • Global Opportunity Lab at UC Berkeley
    Call for Nominations: ELLIS Fellows & Scholars 2025

    Call for Nominations: ELLIS Fellows & Scholars 2025

    The European Laboratory for Learning and Intelligent Systems (ELLIS) is accepting nominations for new Fellows and Scholars

    Inspired by the Canadian CIFAR network (formerly LMB CIFAR), the ELLIS Fellowship Program brings together outstanding researchers through collaborative groups that focus on key research themes via workshops and ongoing collaboration.

    This initiative serves two key purposes:

    1. Recognising and supporting top-tier researchers who are advancing the frontiers of machine learning and AI.
    2. Building a strong, connected European AI community that can influence strategic directions—similar to the Canadian experience, which led to the creation of Mila (Montreal), Vector (Toronto), and Amii (Edmonton) under a national AI strategy.

    About ELLIS Fellows and Scholars

    ELLIS Fellows are outstanding senior researchers—typically with more than ten years of post-PhD experience—who play a key role in shaping the scientific and strategic vision of ELLIS. They act as ambassadors, provide leadership, and actively  contribute to strengthening the ELLIS ecosystem.

    ELLIS Scholars are exceptional early-career researchers—often assistant professors—with up to ten years of post-PhD experience, who are clearly on a trajectory towards becoming future Fellows.

    For ease of reference, the term “Fellow” is used below to encompass both Fellows and Scholars.

    Key Contributions of Fellows

    Beyond research excellence and community-building, ELLIS Fellows contribute to the network in several key ways:

    • Supervising students in the ELLIS PhD program
    • Supporting the establishment of the  ELLIS Units and Institutes
    • Providing strategic input on the growth and direction of ELLIS
    • Participating in the selection process for new Fellows

    Fellowship Programs unite researchers through regular meetings held across Europe, encouraging the exchange of ideas, strengthening community ties, and enhancing the international visibility of European AI research. 

    These programs are currently supported by the ELIAS project, with fellowships guaranteed until the end of ELIAS funding (2027).

    Nomination Process

    Selection is based on:
    • scientific excellence
    • societal impact, 
    • relationship to ELLIS

    Only ELLIS Program and Unit Directors are eligible to nominate candidates for Fellowship.

    Nominations are evaluated by current Fellows who:
    • Are from different countries than the candidate
    • Hold different nationalities
    • Have no conflict of interest
    Z
    Evaluation Criteria (Not Exhaustive)
    • Strong publication record in leading conferences and journals
    • High citation impact
    • International awards and recognitions

    Detailed guidelines on the evaluation criteria were recently updated and can be found here .

      How to Apply

      Interested researchers should reach out directly to the relevant Director.

      🗓️ Deadline for nominations: September 30th, 2025

      More details and the full nomination guidelines are available at:🔗 Open Call for ELLIS Fellows and Scholar nominations 2025

      View last year’s call and selected Fellows & Scholars: 🔗 Fellows & Scholar

      Have Questions?

      For further information or queries, please contact:

      Giulia Clerici

      📧 admin-fellows@ellis.eu