Periodic Reporting for period 1 - ELSA (European Lighthouse on Secure and Safe AI)
Periodo di rendicontazione: 2022-09-01 al 2024-02-29
ELSA, a virtual center of excellence, leads foundational research in safe and secure artificial intelligence (AI) methodology. It collaborates with a growing network of European AI experts to drive cutting-edge AI solutions, positioning Europe as a global AI beacon. Building upon the renowned ELLIS network, ELSA focuses on three pillars: selected research programs, local research units, and a PhD/postdoc program. ELSA has established “The ELSA Innovation Lab” to foster academia-industry collaboration and bridge theory with real-world applications. This initiative is crucial as AI impacts sensitive domains like healthcare and autonomous driving. ELSA tackles three grand challenges: technical robustness and safety, privacy-preserving techniques and infrastructures, human agency and oversight. ELSA has developed its Strategic Research Agenda (SRA) to outline research programs that cut across the three grand challenges, encompassing six distinct use-cases. The SRA represents the ELSA’s guiding document for scientific excellence and rigorous methodologies to address the use-cases in a sustainable manner. Nonetheless, it remains subject to updates as research progresses. ELSA prioritizes machine learning, particularly deep learning, as the bedrock for modern AI applications. The lighthouse aims to harmonize robust technical methods with ethical principles and effective governance structures, fostering AI development aligned with European values. ELSA’s three core objectives are:
Fostering a Unified European AI Community
Advancing Secure and Safe AI
Connecting Research, Innovation, and Deployment
Fostering a Unified European AI Community
Advancing Secure and Safe AI
Connecting Research, Innovation, and Deployment
WP1: The progress on this WP1 is in line with the plan. Within WP1, we have made various contributions to strengthen the secure use of AI. We have contributed to making testing methodologies more reliable and scalable. We have created certifiable mechanisms that are more efficient and applicable to different systems. In parallel, we have developed new techniques to quantify the uncertainty associated with the predictions and tested their resilience against attacks. Furthermore, we have developed new techniques to build machine learning models considering their accuracy, explainability, fairness, robustness, and privacy.
WP2: The work of WP2 has started as planned with the definition of the methodological grand challenge and helping the use cases set up their competitions related to WP2. Research has progressed as planned and made significant contributions to private and collaborative learning. Changes in the broader research environment have led us to consider additional themes in privacy with foundation models as well as private synthetic data generation.
WP3: Our work on human-in-the-loop decision making addressed several aspects. We defined the grand challenge with the community, identified social practices and technical assumptions in AI certification through workshops and interviews, and summarized this in our analysis of AI governance regimes. We investigated embedding technical methods into legal and ethical governance regimes for AI design and deployment. We developed methods for creating interpretable AI systems using prototypes and visualizing the decision-making process. Lastly, we engaged with impactful real-world problems.
WP4: A WP4 workshop was organized to define the framework for benchmarks and specifications of 6 use cases, aligning with the research plans of WP1, WP2, and WP3. A common platform was established for hosting all benchmarks, allowing data download, submissions, and result visualization. The platform, functional since June 2023, provides access to all use case resources and enables participation in ELSA competitions. All use cases have published datasets and metrics on the platform. Networking activities, thematic workshops, and research competitions are underway, with three workshops and one competition being organized.
WP5 and WP6: We progressing as planned. ELSA’s visual identity and communication guidelines established. An open call was launched to finance SMEs, selecting six from 81 applications for collaboration on specific use-cases. ELSA collaborates with ELLIS and ICT-48+ NoEs, and has initiated the ELLIS Lighthouses Committee. The first version of the Strategic Research Agenda (SRA) has been published, with ongoing refinements through collaboration with ELSA’s boards. ELSA covers travel costs for researchers and competition winners, holds regular meetings for management, and uses the ELSA Dashboard for budget tracking and activity monitoring. Ethics, Industrial, and Scientific Boards have been established, with the Executive Board meeting them on demand.
WP2: The work of WP2 has started as planned with the definition of the methodological grand challenge and helping the use cases set up their competitions related to WP2. Research has progressed as planned and made significant contributions to private and collaborative learning. Changes in the broader research environment have led us to consider additional themes in privacy with foundation models as well as private synthetic data generation.
WP3: Our work on human-in-the-loop decision making addressed several aspects. We defined the grand challenge with the community, identified social practices and technical assumptions in AI certification through workshops and interviews, and summarized this in our analysis of AI governance regimes. We investigated embedding technical methods into legal and ethical governance regimes for AI design and deployment. We developed methods for creating interpretable AI systems using prototypes and visualizing the decision-making process. Lastly, we engaged with impactful real-world problems.
WP4: A WP4 workshop was organized to define the framework for benchmarks and specifications of 6 use cases, aligning with the research plans of WP1, WP2, and WP3. A common platform was established for hosting all benchmarks, allowing data download, submissions, and result visualization. The platform, functional since June 2023, provides access to all use case resources and enables participation in ELSA competitions. All use cases have published datasets and metrics on the platform. Networking activities, thematic workshops, and research competitions are underway, with three workshops and one competition being organized.
WP5 and WP6: We progressing as planned. ELSA’s visual identity and communication guidelines established. An open call was launched to finance SMEs, selecting six from 81 applications for collaboration on specific use-cases. ELSA collaborates with ELLIS and ICT-48+ NoEs, and has initiated the ELLIS Lighthouses Committee. The first version of the Strategic Research Agenda (SRA) has been published, with ongoing refinements through collaboration with ELSA’s boards. ELSA covers travel costs for researchers and competition winners, holds regular meetings for management, and uses the ELSA Dashboard for budget tracking and activity monitoring. Ethics, Industrial, and Scientific Boards have been established, with the Executive Board meeting them on demand.
The ELSA consortium has authored more than 120 open-access publications on safe and secure artificial intelligence during its initial reporting period. These publications spearhead cutting edge research including emerging topics and offer solutions to critical challenges. Additionally, ELSA aims to enhance its network, fostering stronger internal and external collaborations. Our unwavering commitment lies in upholding and promoting the EU’s values within the technology landscape, while maintaining rigorous standards of excellence in research. Furthermore, ELSA with its ongoing expansion and numbers events and workshop aggregates and disseminates insights at scale from research communities and society - well beyond our founding consortium. As we continue carrying out this research and activities, we recognize the urgency of today and have and will weigh in on key issues of safe and secure AI as a Europea think tank e.g. by engaging with the upcoming EU AI office and other entities to meet requirements. In order to bring this momentum to bear and to leverage it for Europe, the network should be extended by at least one additional year.