The following text has been machine translated from the German with no human editing.

Large language models (LLMs) – the technology behind tools such as ChatGPT and Claude, which millions of people use every day – only demonstrate their full potential after extensive fine-tuning. This step, known in the field as fine-tuning or post-training, specialises a model for new subject areas, languages or tasks and ensures that it is safer and better tailored to human requirements. Yet although almost every practical application relies on this fine-tuning, surprisingly little is known so far about when, why and how it actually works. This is precisely where Marius Mosbach’s research comes in.

His aim is to transform the adaptation of language models from a collection of empirical recipes into a well-founded science, and to use this understanding to develop more reliable and adaptable AI systems. His work is divided into three areas: Firstly, he focuses on the interpretability of models – that is, the question of how language models work internally and how adaptation changes them. However, Marius Mosbach is also interested in the topic of generalisation – that is, the question of when and why models function reliably even outside their training data. He is also exploring the field of continuous learning, in which systems continue to learn dynamically even after training has been completed, can update outdated knowledge and improve whilst in use. A particular priority for him is to make research into the interpretability of AI systems practically applicable.

Marius Mosbach has made a name for himself in academic circles through, amongst other things, LLM2Vec, a method that enables large language models to be transformed into powerful text encoders. Furthermore, his ground-breaking work on the stability of fine-tuning language models such as ‘Bert’ has attracted widespread attention.

At Saarland University, Marius Mosbach is establishing a new research group within the Department of Language Science and Technology at the Saarland Informatics Campus. At the same time, he will serve as Scientific Director at the German Research Center for Artificial Intelligence (DFKI), thereby maintaining close links with the international research community and AI research at the centre.

Marius Mosbach has received numerous awards for his work, together with his co-authors, including a Best Paper Award at COLING 2022, the Best Theme Paper Award at ACL 2023 and the Most Interesting Paper Award at the BabyLM Challenge 2023. Marius Mosbach will take up his post at Saarland University on 1 October 2026.

Short biography

Marius Mosbach studied Computer Science at Saarland University, where he also completed his Ph.D. at the Faculty of Mathematics and Computer Science. He subsequently conducted postdoctoral research at Mila – Quebec AI Institute and at McGill University in Montréal (Canada). Marius Mosbach will take up his post as a professor at Saarland University and as Scientific Director at the German Research Center for Artificial Intelligence (DFKI) on 1 October.

The German Research Center for Artificial Intelligence (DFKI) and the Canadian research institute Mila have agreed to join forces in key sectors to strengthen German-Canadian AI ties. The partnership was formalized at ALL IN, Canada’s largest AI and technology event, in Montréal, where the leaders of both institutions signed a Memorandum of Understanding in the presence of German Federal Minister for Digital Transformation and Government Modernisation Karsten Wildberger and Canadian Minister of Artificial Intelligence and Digital Innovation Evan Solomon. Building on recent government efforts to strengthen bilateral ties in applied AI and research, the collaboration focuses on joint research projects in key application domains such as the energy sector, as well as the development of trustworthy and safe AI systems and agentic AI.

Synergies for the Energy Transition and AI Safety

Both Germany and Canada face the challenge of transforming their energy systems into sustainable, efficient, and resilient structures. Artificial intelligence plays a key role here—for instance, in optimizing smart grids, forecasting renewable energy generation, and improving industrial energy efficiency.

The partnership between DFKI and Mila combines the complementary strengths of two leading global AI ecosystems. In addition to solutions for the energy transition, the partnership focuses on research questions surrounding agentic AI, AI safety, and trustworthy systems. Through targeted knowledge sharing, methodology exchange, and talent mobility, the partners aim to develop practical solutions and accelerate technology transfer into the Canadian and German economies.

Statements on the Cooperation

“The cooperation with Mila marks an important milestone for DFKI. Canada and Germany share the vision of human-centric and trustworthy artificial intelligence. By combining our research and transfer capabilities, particularly in critical application domains such as the energy sector, we create synergies that benefit science, industry, and society in Canada and Germany alike.”

Prof. Dr. Antonio Krüger, CEO of DFKI

“As Mila continues to expand its global reach, joining forces with one of Germany’s leading AI research institutes is an important next step. As a trusted partner, Germany, and more specifically DFKI, will enable us to advance cutting-edge AI research and translate it into high-impact industrial applications.”

Valérie Pisano, President and CEO of Mila

“This partnership brings together two of the world’s leading AI research institutions with the clear goal to further develop safe and trustworthy AI systems. This collaboration not only accelerates AI innovation, but also ensures that artificial intelligence serves people, businesses, and society in Germany and Canada.”

Dr. Karsten Wildberger, Federal Minister for Digital Transformation and Government Modernisation, Federal Republic of Germany

“In a rapidly evolving AI landscape, progress depends on trusted partnerships. By bringing together two world-leading AI institutions, Mila and DFKI, Canada and Germany are strengthening the ties between our research ecosystems and advancing the development of safe, reliable AI. This collaboration will help accelerate innovation, deepen scientific exchange, and ensure that AI serves people, businesses, and societies on both sides of the Atlantic.”

The Honourable Evan Solomon, Minister of Artificial Intelligence and Digital Innovation and Minister responsible for the Federal Economic Development Agency for Southern Ontario, Government of Canada

About Mila

Founded by Professor Yoshua Bengio, Mila is one of the world’s leading AI research institutes, bringing together close to 2,000 researchers & professionals shaping the future of intelligence. A non-profit organization based in Montreal, Mila is recognized for its scientific contributions, global innovation partnerships, leadership in safe & responsible AI, and acceleration of AI venture creation. It was created through a unique partnership between Université de Montréal and McGill University to advance scientific breakthroughs for the benefit of all. Mila is supported by the Government of Canada, the Government of Quebec, and more than 150 industrial partners. For more information, visit mila.quebec. 

About DFKI

The German Research Center for Artificial Intelligence GmbH (DFKI) was founded in 1988 as a non-profit Public-Private Partnership (PPP). It operates facilities in Kaiserslautern, Saarbrücken, Bremen, Lower Saxony, and Darmstadt, labs in Berlin and Lübeck, as well as a branch office in Trier.

DFKI combines scientific excellence and commercially oriented value creation with societal impact. For over 35 years, DFKI has been conducting research on human-centric AI, guided by social relevance and scientific excellence across key future-oriented research and application domains of artificial intelligence. In the international scientific community, DFKI is recognized as one of the premier “Centers of Excellence.” Currently, approximately 1,400 employees from over 76 nations are researching innovative software solutions.

The legal IT specialist and his team have developed tools for specific use cases, which they have bundled—along with instructions—in the free and openly available “Legal Work with AI” toolbox—ranging from brainstorming and research to text review.

The following text has been machine translated from the German with no human editing.

In legal work, there are numerous tasks that take a great deal of time. One example is searching for references. “Research has traditionally accounted for a significant portion of working hours when dealing with legal texts,” explains Professor Georg Borges, Executive Director of the Institute for Legal Informatics. Borges and his team have developed an automated reference search tool for this purpose. “In our experience, this can save 70 to 80 percent of working time. Using the latest AI models, we’ve achieved a real breakthrough here,” says the legal informatics expert, citing an example: “I was able to automatically insert 200 individual citations into my most recent essay in one hour; verifying them took three days.”

The Saarbrücken Law Toolbox currently includes a dozen such applications for drafting legal texts for research and academic study. “For future lawyers, the professional use of AI tools is a key skill,” says Borges. The toolbox makes it easier to use them.

“The tools we’ve developed address tasks that are central to everyday legal work. In academic writing, for example, AI helps organize thoughts, improve outlines, and refine drafts. As a critical discussion partner, it can also help scrutinize arguments and identify potential weaknesses in legal reasoning,” explains the legal informatics expert.

The toolbox also includes useful tools for studying and teaching: “Here, use cases range from digital index cards to transcribing lectures and interviews. This also includes creating presentation slides based on sample legal solutions and condensing lengthy slide decks,” says Georg Borges. The AI-powered citation search is particularly interesting in this context as well. “Anyone writing an essay, a dissertation, or a term paper needs sources to support specific claims. Especially when dealing with specialized topics, finding a suitable passage can be time-consuming,” says Borges. AI should not merely suggest literature on the same topic, but rather identify specific sources. References to opposing viewpoints are also valuable for critically evaluating one’s own arguments.

“The AI services can be linked to researchers’ own collections of specialized literature. This allows researchers to build on materials they have already compiled for their work,” explains the lawyer. This presents an opportunity to significantly streamline time-consuming search work and create more space for in-depth analysis of the content. Borges and his team have developed additional AI tools for analyzing formal text errors, verifying citations, and checking bibliographies for potentially fabricated sources.

Borges emphasizes that what is important in all of this is that humans must always remain in control of the process and understand their role in this context: “Responsibility for the use of AI always rests with humans: It must always be verified by an expert whether, for example, a reference is reliable and whether the original source actually supports a claim,” says Georg Borges.

The free and openly available toolbox will be presented in an interactive format at the Institute for Legal Informatics’ booth during the EDV-Gerichtstag, September 23–25, 2026, in Building B4 1 at Saarland University.

You can find the press release for the IT Law Conference itself at the following web link: https://www.uni-saarland.de/aktuell/edvgt-2026-50529.html

For more information, please visit the Institute for Legal Informatics’ website:
https://www.rechtsinformatik.saarland

You can find the press release for the IT Law Conference itself at the following web link: https://www.uni-saarland.de/aktuell/edvgt-2026-50529.html

About the Institute for Legal Informatics

The Institute for Legal Informatics at Saarland University is one of Germany’s leading research institutions in its field. It conducts research and teaches on the legal and technical issues of the digital world. Topics include IT security, data protection, electronic court proceedings, digital administration, smart energy systems, and issues related to autonomous systems. In addition, the institute is committed to training the next generation of scholars. This includes specialized courses, certificate programs, and a master’s program in information technology and law.

Questions answered by:

Prof. Dr. Georg Borges, Institute for Legal Informatics, Saarland University
Phone: +49 (0)681 302-3105 ; E-Mail: georg.borges(at)uni-saarland.de

Whether it is novel digital models for representing the world or the development of new cloud infrastructures for artificial intelligence, the European Research Council (ERC) funds excellent pioneering research through Starting Grants. These grants support early-career researchers worldwide whose work has the potential to shape our society. This year, two members of the Max Planck Institute for Informatics are being awarded a Starting Grant: Dr. Yiting Xia and Dr. Thomas Leimkühler will each receive around 1.5 million euros to carry out their projects over the next five years.

Dr. Yiting Xia is a Tenure-Track Faculty member at MPI-INF and heads the Network and Cloud Systems research group. In her ERC project, Synchronous Optical Network Infrastructure for the Cloud (SONIC), she investigates how data centers can be efficiently adapted to meet the demands of AI workloads.

The focus of her project is on Data Center Networks (DCNs), the critical network infrastructure for cloud computing. These serve as the nervous system of a cloud data center, functioning like a fully automated sorting system in a massive logistics hub: they direct a digital flood of data packets to their correct destinations, determining data paths in fractions of a second. Traditionally, electrical packet switches have handled this task—acting as traffic crossings to forward each individual data packet. However, the boom in artificial intelligence is pushing these electrical switches to their physical limits. The enormous data volumes of modern AI models overwhelm the hardware, leading to delays in packet delivery and a massive increase in energy consumption, ultimately preventing the infrastructure from scaling effectively.

Yiting Xia aims to solve this problem by rethinking DCN infrastructure: moving away from electrical data transmission toward optical networking technologies, which transmit data streams as light signals. She plans to leverage powerful optical components, such as Optical Circuit Switches, to transform DCNs from a mere connectivity substrate into active system elements for AI.

Training and serving large AI models require thousands to hundreds of thousands of accelerators (such as GPUs) performing highly coordinated, synchronous computations. “These processes exhibit recurring and predictable communication patterns. In my project, I am investigating whether the network can be redesigned around the structure of AI computation, rather than forcing AI workloads onto networks designed for traditional, unpredictable traffic,“ says the Saarbrücken-based Max Planck researcher.

The goal is to significantly increase the data-delivery and energy efficiency of DCNs through new technologies and close coordination of system components, while simultaneously enhancing resilience and simplifying maintenance.

Dr. Thomas Leimkühler is a Senior Researcher at MPI-INF, where he leads the Image Synthesis and Machine Learning research group. In his ERC project, From Microscopic to Monumental: Learning Visual Scene Representations Across Vastly Different Scales (MonuMicro)“, he aims to solve a fundamental challenge in visual computing: How can our world be digitally represented in such a way that both entire landscapes and microscopic details are unified in a single model?

Modern reconstruction and rendering techniques can, for example, capture individual objects with high precision or realistically depict entire cities and landscapes. However, when vastly different scales come together, they reach their limits: A model describing a mountain landscape loses minute details, while a model of an insect can no longer relate to its surroundings. This is where Leimkühler’s MonuMicro project comes in. Instead of incrementally improving existing methods, he seeks to establish entirely new theoretical and algorithmic foundations to represent the visual world across many orders of magnitude in a single, cohesive model.

To achieve this, MonuMicro combines the mathematical scale-space theory with modern machine learning and generative AI methods. The new approach explores, for instance, how missing information can be inferred from a few recorded image data points while ensuring that fine details and large-scale structures remain consistent with one another.

“The results could be applicable across many fields: In robotics, autonomous systems could gain a much better understanding of their surroundings; virtual and augmented reality could become more realistic and experience no noticeable loss of detail. Other sciences, such as archaeology or geosciences, could also benefit from the ability to visually connect large contexts with the finest details,” says the Saarbrücken-based Max Planck researcher.

Thomas Leimkühler has experience in developing new scene representations: 3D Gaussian Splatting, a technology co-developed by Leimkühler, has become the international standard for efficiently rendering three-dimensional scenes and has been cited nearly 12,000 times in just three years. With this ERC project, he is now taking a decisive step further, aiming to lay the groundwork for representing the visual world across many scales in a unified model.

According to the European Research Council, a record 4,807 research proposals were submitted in the current round of ERC Starting Grants. Of these, 421 were selected for funding (8.8%), totaling 705 million Euro. Across Europe, 22 projects in computer science received funding, five of which are from Germany. Two of these five projects are based at MPI-INF, with a third going to Michael Hahn from Saarland University. This means that 2026 three of the five German ERC Starting Grants in computer science are located at the Saarland Informatics Campus.

Further Information:
ERC press release: https://erc.europa.eu/news-events/news/erc-2026-starting-grants-results
Website MPI for Informatics: https://www.mpi-inf.mpg.de/de/home
Website of Yiting Xia’s research group: www.mpi-inf.mpg.de/de/departments/network-and-cloud-systems
Website of Thomas Leimkühler’s research group: https://ismael.mpi-inf.mpg.de/

Editor and press contact:
Philipp Zapf-Schramm
Joint Administration
Max Planck Institute for Informatics
Max Planck Institute for Software Systems
Tel: +49 681 9325 4509
Email: pzs@mpi-klsb.mpg.de
 

He has now been awarded a European Research Council (ERC) Starting Grant worth €1.5 million over five years to pursue this work.

Large language models are being used to handle increasingly complex tasks to which they are not yet particularly well suited. Problems arise when companies use these models to manage complex planning scenarios such as coordinating staff rotas with material deliveries, production processes and delivery deadlines. ‘Language models still struggle with sequences of interdependent events, especially when the relevant information comes from different sources. When drawing up a staffing rota, for example, a model must take account of when materials will arrive, which production steps depend on them and whether enough staff are available to meet the delivery deadline. We want to understand how the theoretical foundations of language models need to be improved so that they can keep track of such interdependent sequences more reliably,’ says Michael Hahn, Professor of Computational Linguistics at Saarland University. Using large datasets, he and his team want to investigate which relationships AI systems can infer successfully from the data and identify those cases where they still fall short.

‘Large language models find it particularly difficult to distinguish between very similar pieces of information. This is the case, for example, in biomedical research, where descriptions of molecules and cellular processes may differ only in subtle respects,’ explains Michael Hahn. Hahn wants to develop a deeper scientific understanding of how AI systems learn and how they arrive at conclusions step by step. ‘Inductive bias plays a key role,’ says Hahn. ‘This is the set of assumptions and prior knowledge that a machine-learning algorithm needs in order to be able to generalize from familiar training data to data it has not encountered before.’ 

He and his team will examine the underlying mathematical models in depth and plan to develop a theory to explain how different training conditions affect the ability of an AI system to develop the low-level and high-level reasoning required to reach a well-founded conclusion. ‘We want to develop the theory using real-world tasks so that we can both predict and prevent model errors. Our goal is to make language models more reliable and cost-effective and, ultimately, help pave the way for trustworthy AI,’ says Hahn.

Michael Hahn has now been awarded an ERC Starting Grant worth €1.5 million over five years for his project ‘REALM: Foundations for Reliable Language Model Reasoning via Inductive Biases’. Last year, he received a similar amount through the German Research Foundation’s Emmy Noether Programme to gain a better understanding of the fundamental architecture of language models and explore new approaches to their design (see press release of 19 November 2025). In March, Michael Hahn was also awarded the Heinz Maier-Leibnitz Prize. 

Computational linguist Michael Hahn is a member of Saarland University’s Department of Language Science and Technology, which conducts internationally renowned research and teaching at the intersection of language, cognition and artificial intelligence. The department maintains close links with the Saarland Informatics Campus. In the latest funding round, Yiting Xia and Thomas Leimkühler of the Max Planck Institute for Informatics also received ERC Starting Grants. This brings the number of researchers at the Saarland Informatics Campus who have secured one of the European Union’s various ERC grants to 54. Of the 421 ERC Starting Grants awarded in the current round, 22 are in the field of computer science. Only five of these went to Germany, three of them to the researchers at the Saarland Informatics Campus named above.

Background information – Saarland Informatics Campus 

One thousand scientists and about 2,800 students from more than 81 nations make the Saarland Informatics Campus (SIC) one of the leading locations for computer science in Germany and Europe. Four world-renowned research institutes, namely the German Research Center for Artificial Intelligence (DFKI), the Max Planck Institute for Informatics, the Max Planck Institute for Software Systems, the Center for Bioinformatics as well as Saarland University with three departments and 24 degree programs cover the entire spectrum of computer science.

Further information:

ERC Starting Grants 2026: https://erc.europa.eu/news-events/news/erc-2026-starting-grants-results
Department of Language Science and Technology at Saarland University
Professor Michael Hahn’s personal website: https://www.mhahn.info

Questions can be addressed to:

Professor Michael Hahn 
Language, Computation and Cognition Lab
Tel. +49 681 302-4343
Email: mhahn(at)lst.uni-saarland.de

To prepare graduating students for professional life and provide insights into various career fields, so-called career information days are an integral part of the curriculum at many schools in Saarland – including the Otto-Hahn-Gymnasium in Saarbrücken. Our Linguistics and Language Technology track was also a guest at this year’s event –represented by student assistants Léon Jost and Elena Dort, as well as Rainer Egler, an external lecturer. All three are, incidentally, alumni of the Otto-Hahn-Gymnasium themselves. In two sessions, they provided a total of 23 interested students from the modern languages track with information about the content and structure of the bachelor’s program in Language Science. Topics included the various fields of linguistics, the practical focus of the program, experiences from a student’s perspective, and potential career prospects after completing the bachelor’s degree. Léon and Elena, both current or former Language Science students at Saarland University, shared firsthand accounts of their studies and student life, while Rainer introduced the department and the various fields of work. Following the presentations, the students had the opportunity to ask questions.
All in all, the career information day provided a wonderful opportunity to introduce the B.A. Language Science to prospective students and offer a personal glimpse into what it is like to study at Saarland University. We are excited to see whether some of these students will indeed find their way to us, and we look forward to future events of this kind!

The ACM SIGPLAN Most Influential ICFP Paper Award is a retrospective award—it is given each year to the paper deemed most influential from the ICFP conference 10 years earlier.

A video of the award presentation can be found here: https://www.youtube.com/live/wn88R35yqeY?t=26630s

Award citation: Higher-Order Ghost State represents a major milestone for the widely used Iris program logic framework.  Colloquially known as “Iris 2.0”, the paper contributes a key feature: the ability to store arbitrary higher-order separation logic predicates in ghost variables.  The feature is justified using guarded recursion and a new algebraic structure called cameras – roughly, a generalization of PCMs with a form of step indexing.  Higher-order ghost state was later shown to be an essential building block from which other features like impredicative invariants and weakest preconditions could be derived.  Its broad and lasting impact is evident in the (currently) 165 papers that use Iris in some way, including 45 POPL, 29 PLDI, 25 ICFP, and 20 OOPSLA papers.

Businesses, public authorities and representative organizations are invited to attend the kick-off event free of charge and to contribute topics of their own. 

The following text has been machine translated from the German with no human editing.

When a damaged area of woodland is replanted in the Saarland, it takes years to see whether the young trees are taking root. Until now, this has only been visible on satellite images once a canopy has formed, often after five years. Angela Nyangate John has developed a method that detects reforestation around a year after planting. In a test area in Kenya, it identified 94 per cent of the sites where forest later grew, whilst a standard method detected only 47 per cent. Worldwide, several hundred million hectares of forest are set to regrow by 2030. This method would make it possible to assess whether this is successful much earlier. Partners from the Saarland forestry administration are registered to attend the event on 2 September and will take part in the discussions.

Traces that arise in everyday life anyway are increasingly replacing surveys, counts and site inspections. Satellite images show how heavily a location is visited, even if individual cars are not visible in the images. Theophilus Aidoo and Till Koebe analyse this data from car parks. Barriers and pay-and-display machines provide such figures for individual car parks, whilst satellite images provide them for all car parks in a region simultaneously. A retailer can use this to gauge how busy an area is beyond their own door, or to identify where in a city there is a shortage of parking spaces and where they remain empty.

Google reviews of swimming pools reveal what bathers write about their encounters with other visitors. Ethel Mensah has analysed 1.2 million of these. Around 40 per cent of the reviews concern such encounters; in roughly two per cent of these, the people involved are described in terms of their origin, ethnicity, culture or religion – and this proportion is rising. Whether there are links to the local political and social situation is part of the ongoing investigation.

These examples represent over a dozen projects from which applications for society, the economy and public administration in the Saarland may emerge.

SOUNDS kick-off event on 2 September

The kick-off event on 2 September begins at 9.30 am and has two main focuses: research and its methods in the morning, and practical applications in the afternoon. Melanie Revilla from the Institut Barcelona d’Estudis Internacionals will report on what six years of research using image, language and usage data have revealed about the reliability of such sources. Afterwards, the researchers will present their projects and discuss the findings at their research posters. Both sides will then seek common ground in short discussion sessions.

In the afternoon, Christina Elmer, who holds Germany’s first professorship in data journalism at TU Dortmund University, will demonstrate how newsrooms close data gaps using data donations, proxy variables and their own measurements. Afterwards, the guests will split into three workshops. Together with East Side Fab, the innovation network in the Saarland, experts from business, public administration and civil society will work with the researchers to develop joint research questions and initial projects. Two workshops with Triathlon, the university’s start-up and innovation centre, will focus on spin-offs – a career path that is rarely considered by early-career researchers. University President Ludger Santen, Lord Mayor Uwe Conradt and Minister for Science Jakob von Weizsäcker will deliver the welcoming addresses.

Register at:  https://www.uni-saarland.de/forschen/sounds/kick-off-veranstaltung.html

Background: The SOUNDS transformation project

The Societal Observatory Using Novel Data Sources (SOUNDS) is based at Saarland University. It is led by Daniela Braun, Professor of Political Science specialising in European integration and international relations, and Ingmar Weber, Alexander von Humboldt Professor of Artificial Intelligence specialising in societal computing. SOUNDS takes socially relevant questions as its starting point and answers them using methods from Computer Science and novel data sources, including satellite imagery, social media posts, search queries, reviews and mapping services. The findings are intended to support decision-making, promote public participation and make democracy more resilient. 

The state government has been funding SOUNDS since 2025 with 29 million euros from the Saar Transformation Fund for Research and Knowledge Transfer; the project will run until August 2032. This funding is intended to create jobs, foster collaborations with businesses and public authorities, and lead to spin-offs from the research.

Further information: www.uni-saarland.de/sounds

Media representatives are invited to the kick-off event and should also register via the following link. The project team can arrange interview slots with the researchers and provide photographic material: projekt-sounds(at)uni-saarland.de

Saarland University is represented in the project by Professor Christoph Sorge and his team at the Chair of Legal Informatics.

The following text has been machine translated from the German with no human editing.

The following press release has been issued by the Fraunhofer Institute for Experimental Software Engineering (IESE) in its capacity as consortium leader.

A protocol for greater digital sovereignty

Managing privacy settings poses significant challenges for many people. Consents and preferences are currently spread across a multitude of digital services, whilst existing solutions are often not very user-friendly. At the heart of the ddpp project is the development of a decentralised web protocol for data protection preferences. Instead of having to manage data protection settings separately within each individual service, users will in future be able to control their preferences via a so-called ‘Personal Privacy Preferences Place’ from a central location. The preferences are then automatically synchronised with the connected services. The aim is to strengthen privacy-friendliness, transparency and digital sovereignty, whilst at the same time creating a standardised, interoperable solution that can be integrated into existing digital services.

The project is based on the ‘Privacy by Default’ principle: privacy-friendly settings are to be technically supported from the outset, so that users can effectively protect their privacy without having to navigate complex settings or numerous consent requests. The project results are also intended to form the basis for potential standardisation and to introduce key concepts to relevant expert and standardisation bodies.

‘Many people today lose track of how their data is being used by various digital services. With ddpp, we are developing an approach designed to significantly simplify the management of such settings. If we succeed in establishing the protocol on a broad scale, it could give rise to a new standard for handling data protection preferences,” says Dr Frank Elberzhager, Head of the Architecture-Centric Engineering Department and project leader at Fraunhofer IESE.

Concrete benefits in key application areas

The relevance of standardised management of data protection preferences is particularly evident in two areas of application being investigated in the project. In the field of personalised advertising, users should not have to make decisions about cookies and data usage anew for every digital service. Instead, their data protection preferences could be set once and automatically applied across all services. This would reduce the need for repeated consent requests and strengthen users’ control over their own data.

A second use case involves health data. Here, a wide range of digital applications and connected devices process particularly sensitive information. The project is investigating how users can determine for themselves, even within such ecosystems, which health data they wish to share with which applications, and how these decisions can be implemented.

Technology and law considered together

Fraunhofer IESE is responsible for the technical design, development and evaluation of the protocol. This includes, amongst other things, the development of the protocol architecture, the prototyping of key components, user-centred interaction concepts, and the analysis of market and roll-out strategies. Saarland University is contributing its expertise in the fields of legal informatics, data protection law and regulatory frameworks. Together, the consortium is pursuing an interdisciplinary approach that brings together technical, legal and user-centred perspectives.

About the ddpp project

The ddpp (Decentralised Data Privacy Protocol) research project is developing an open, decentralised web protocol for the cross-service management of data privacy preferences. The aim is to establish a standardised and interoperable solution that simplifies data protection decisions for users whilst supporting organisations in implementing regulatory requirements in a manner that complies with data protection legislation. The project runs from 1 July 2026 to 30 June 2029. The project partners are the Fraunhofer Institute for Experimental Software Engineering IESE (consortium lead) and Saarland University.

The project is funded by the Federal Ministry of Research, Technology and Space (BMFTR) as part of the funding scheme ‘Privacy Platform – IT Security Protects Privacy and Supports Democracy’. It contributes to strengthening privacy, informational self-determination and trustworthy digital infrastructures.

Academic contacts:

Professor Christoph Sorge (Saarland University): Email: christoph.sorge(at)uni-saarland.de , Tel.: +49 681 302 5120 

Fabienne Bäcker (Fraunhofer IESE), email: fabienne.baecker(at)iese.fraunhofer.de

Press photos available for download: You may use the press photos free of charge in connection with this press release and reporting on Saarland University, provided you credit the photographer.
 

Currently, she’s a research scientist at Meta. Before that, she was a postdoctoral researcher in the Computer Vision and Machine Learning department at the Max Planck Institute for Informatics (MPI-INF).

After being exposed to computer vision during her bachelor’s in Russia by coincidence, Anna decided to stay in the field throughout her entire education. Upon completing her PhD in Saarbrücken, she was awarded the Otto Hahn Medal for her thesis “Advancing Image and Video Recognition with Less Supervision,” in which she explored how to get good results on unseen data with minimal labeling.

In this interview, Anna talks about her award-winning thesis, her experience as a researcher at MPI-INF and as a woman in computer vision, and her advice to aspiring computer scientists.

What drew you to computer science and more specifically, to computer vision? 

It was almost an accident, to be honest. I actually planned to go into mathematics. In Russia, we have an international unified exam, and I was looking at the list of subjects I needed to pass. Russian and mathematics are mandatory, so you don’t need to choose them. In addition to those, I only had physics, while everyone else in my class was choosing more subjects. So I thought, why shouldn’t I? That’s when I decided to take computer science. There was nothing I didn’t like about it — it was almost math, but slightly different, and I found that I could learn it quickly. After getting my exam results, I thought, why not choose computer science as my major? 

When I started my bachelor’s, I was the only one in class who didn’t know how to program. That was slightly alarming, but in the beginning it was mostly math anyway. 

After my first year of studying, I ended up in a summer camp — again, by accident. (Lots of happy accidents!) We worked on a quadrocopter (drone) in four teams, with each team responsible for a different section. For instance, one team developed the app’s web page, while our team was responsible for computer vision. I did camera calibration in C++, which I didn’t know at the time, and it became my task for the entire camp. I had a lot of fun doing it and fell in love with computer vision. I decided to stay in the field, and in the end, it worked out! I spent two years in a computer vision lab during my bachelor’s, and when I moved to Germany for my master’s, I joined another computer vision lab as a HiWi.

I like working with images to understand how computers interpret and extract information from what we see — which is, in memory, actually just numbers representing pixels. It’s intriguing that we can reason more and more just from numbers. This is fascinating. The field is developing so fast. It’s incredible to be part of it and stay on top of how these systems work: what kinds of algorithms they use, how they are trained, and what their limitations are. I wouldn’t necessarily say it’s more interesting to be in the field right now, but it’s certainly different, because it’s applied in the real world more than ever. Right now, computer vision is deep learning, which has increasingly been integrated with language models over the past few years.

Currently, you can just upload images to a model and ask, ‘What is in this image?’ That’s related to my field and what I’m doing, because the model needs to understand the context of the image or video. Developments like these are extremely motivating to stay in the field and keep on pushing the research boundaries. The pace of the field has been crazy in these past few years. Nobody knows what’s coming because there’s always something new happening. You see papers coming in the arXiv every day.

You were awarded the “Otto Hahn Medal of the Max Planck Society” for your thesis, “Advancing Image and Video Recognition with Less Supervision”. What piqued your interest in the topic and can you tell us a bit more about your thesis?

Absolutely! The topic is about how we can learn from data. 

To give you more context, I started my path in computer vision research with unsupervised learning. Imagine you have a bunch of data, a bunch of videos — and you’re trying to understand how coffee is made. There are actions like picking up the cup, putting on the cap, stirring sugar, pouring water, and you want to recognize all these actions — but in a way that doesn’t require supervision in the first place.

Supervision is very hard to obtain. It’s very multifaceted and comes with limitations; you can’t label everything in the world. There are a lot of underrepresented classes, and things change all the time. For example, today we sit on these chairs, and in five years they might look completely different, but their purpose will be the same: we still sit on them. In this case, you would need to re-label and add these new chairs to your model, so it can recognize them. 

There are problems with labeling as well. Going back to the chair example, you can label the whole chair, or just the arm or back, depending on the granularity of your labeling and the downstream task you want to solve. 

What I was primarily interested in was how we can exploit the data in different scenarios. When we talk about video understanding, image understanding, or multi-modal understanding, the question is how we can generalize using only a subset of absolutely unlabeled data. Can we achieve good results on unseen data? On test data? That was my topic, in a nutshell.

Regarding my interest in the topic, I think I found common ground with my supervisor. I wanted to work with less supervision. When I started my PhD, we began with few-shot learning, where you’re given very few examples and need to learn your representations from them — and you then model based on these few samples. 

So we started there, and because I like videos, I steered back to working with video data. Fast-forward to later, I was working with large-scale, unlabeled data, focusing on how to get better representations and improve datasets using LLMs. 

It was always about less supervision, mostly with videos but sometimes with images, and about how different learning dynamics influence the kinds of models we actually get. The goal was to understand the internal structure of algorithms. It was also very interesting to see exactly what a model learns and what influences the structure of its representations. This focus was always there: learning with less data and overall less supervision. And then we explored different directions along these lines.

Your thesis mentions that fully supervised models are often impractical in real-world applications. Which fields or industries do you think struggle the most with this necessity for data labeling?

I think any field would be limited. Especially when you have a pre-defined, structured pipeline, like in car manufacturing. There, you have details that always have the same size and shape. But if you don’t have a window, then the outdoor lighting doesn’t make any difference — you can set up the lighting at a fixed point so everything always looks the same. When that happens, robots or algorithms can manipulate objects without much difficulty, because you can define precisely where and how things should be done. 

So the factors and the objects are fixed. It’s a sealed, perfect environment where nothing changes. Whenever something starts to change, though? You’ve got a problem. Imagine you have a table and your task is, say, recognizing objects to help you make coffee in a VR setup (like those from Apple or Meta). Your coffee-making mechanism needs to recognize the cup, the steamer, the spoon, the coffee itself — essentially, everything has to come together. Afterwards, the system should also suggest the actions you’ll need to take. 

Maybe the system doesn’t work if there’s too much or too little light, so it should be able to tell you to open the window, turn off the light, and so on. Keep in mind that this is still an example with rigid objects.

Let’s think of a plastic bag instead, which changes shape constantly. Imagine filling it with powder or liquid — a common scenario in medical applications. Sounds easy, you just take the bag and fill it, right? No! This is actually very complex because a plastic bag isn’t rigid and is constantly changing. There’s also the durability factor. We have sensations that tell us when the bag is under too much stress, close to ripping, and how much force we’re applying to carry it. Our senses give us that feedback — but robots don’t have that. For them, controlling that is extremely hard. 

Essentially, take any application where you have some degrees of freedom, and you’ll see that it’s very hard to label everything. Take autonomous driving, for example. 

Of course, large teams at every car company create proprietary datasets (which aren’t available to academia because they’re not open source). But still, they can’t label everything, as some things are simply unpredictable and it costs a lot of time and money. 

However, generally speaking, it’s not sustainable for every industry or company to collect its own datasets and guarantee high quality. You often need three annotators per data point, not one. You need to be consistent throughout the entire process, and ensure the annotators do a good job and that there aren’t mistakes involved. It’s very common to have mistakes in datasets, because labeling is hard. So essentially, name the field —and you will definitely need methods that rely on less supervision.

Fundamentally, the best approach that I see is to start with a small dataset with very high-quality labels and then propagate this to unlabeled data. Perhaps it’s not an approach that works absolutely without labels, but for some downstream tasks, some guidance is still needed in terms of what labels are expected and what kind of supervision is required. The model can catch up gradually.

However, fully labeling everything is unreasonable, because the world is constantly changing. Then you would need to constantly update and retrain models. It’s much better and effective if a model can pick up on something new. Models should be able to recognize new concepts and integrate them into their representations over time.

And currently, can this approach be used for complex tasks?

I think so. It depends, because you can’t always take something and apply it to completely different data — it might not work. You need to understand the internal structure of the data and how to apply it. There are also additional biases we have as humans, in terms of how things work and what to expect.

Let’s think about an instructional video on making coffee. You know that you’ll first get a cup and then pour water, not the other way around. One action comes first, we know that. This kind of bias can also be observed in other data structures, especially biological or molecular. There are some connections that should be there and some that shouldn’t. You can explicitly or implicitly impose this on a model while it’s learning, and thereby help it recognize things.

What were some challenges you faced during your research and how did you overcome them? 

Every PhD has certain ups and downs. Most of the time, it’s the moments when you’ve been trying to figure out something and suddenly, nothing works and you have no idea what to do next. There are different ways to overcome this. What worked for me was taking a step back — going for a walk, overall, relaxing.

Sometimes when you constantly try to make a problem work and it still doesn’t — you get lost in small details and can’t think of new solutions. When that happens, you need to do something absolutely different — distract yourself for a bit. Read papers, talk to people, sleep, whatever you do: don’t think about the problem. When you’re done, come back and try something different.

It helps! Sometimes it might take more time — maybe weeks, maybe months — but it happens eventually. This is the main PhD challenge for everyone, I think. Of course it’s always frustrating when something you think will work doesn’t, but part of the process is learning how to deal with this. You will have ups and downs. It doesn’t get easier, to be honest, but you learn how to handle it.

In addition to that, for me, starting my PhD during Covid was a huge challenge. I essentially moved to a new city, knowing no one, only a month before Covid hit. I went to the office for a month, and then it closed down. Socially, that was pretty hard. It’s interesting, really — when you’re frustrated and alone, it only adds more pressure.

I did an internship right at the beginning of my PhD as well. Thinking about it now, I think it was too early. It didn’t work out in the end. My manager left as soon as I joined, the project itself fell apart, and although I had the opportunity to continue, I think I made the right decision to say, ‘Okay, I’m not doing this anymore.’ I feel like it takes some courage to tell your supervisor that. Afterwards, I moved to another project. 

I would say it’s important to know when to step back sometimes. In this particular case, I’m quite proud of myself. 

Before working at Meta, you were a postdoctoral researcher at the Max Planck Institute for Informatics at the Saarland Informatics Campus. What were you working on?

I was involved in several projects. I helped PhD students, supervised master’s students, and also worked on my own research. One of my projects focused on understanding attention in diffusion models, which are generative models that create images or videos, such as DALL·E or Sora. You go to the app or the webpage, send a prompt that goes, ‘I want to have a cute cat,’ and the model generates it.

Diffusion models can be used for many different tasks, from image generation to extracting internal representations for applications such as segmentation. I specifically studied attention layers, which are key building blocks of these models. They are where communication between an image and the model’s internal representations takes place. Conceptually, it’s an operation — like multiplication, but far more complex — and we studied how it functions inside the diffusion models, how it affects image generation, and how it shapes the representations learned by the model. At the time, we were targeting a specific segmentation task. We had found a mechanism that amplifies certain attention regions in the internal representations, improving segmentation performance. Essentially, we discovered how to adjust the attention operator and its outputs without retraining the model.

Although the focus of my research remained on computer vision, we were increasingly incorporating language as well. We explored how images or videos could be combined with language to improve internal representations or enable reasoning about visual scenes. This included tasks such as answering questions about visual content, describing relationships between objects in an image, and incorporating additional modalities like audio, depth estimation, or optical flow.

This research was part of a collaboration with Google. Our group at the Max Planck Institute, together with Christian Theobalt’s group, participated in the VIA collaboration. I worked closely with researchers at Google, met with them regularly, and co-authored research with one of the collaborators.

What set the Max Planck Institute for Informatics and SIC apart for you in terms of research opportunities, as well as academic and professional development?

First off, during my interview with my supervisor, I immediately felt a connection. I think it’s super important to see if things click with your supervisor before starting a new position, so I was really happy that was the case. Our communication felt established from the start. 

Oh, and they have a great coffee machine! I’m telling you, I had an internship in Paris — I didn’t expect this from France, but the coffee machine? Quite bad. At MPI, though? I even learned how to make coffee art during my PhD!

Jokes aside, the more time I spent there, the more I learned about the Computer Vision and Machine Learning department and how it was structured. The infrastructure was impressive compared to any other place I had seen in academia. Not just academically, but also in terms of the clusters and setups we used to train our models. There was a dedicated IT department — handling server management, laptops, devices — so that everything ran smoothly. 

I was surrounded by truly inspiring people and there were plenty of possibilities to learn, explore, and collaborate. In our department, we had subgroups that worked on very different topics, and you can learn a lot from that. If you were interested, you could even collaborate. Overall, it was quite different from other places because we had three or four professors in close collaboration under this department.

We also had retreats twice a year, not just within our group but jointly with other groups across Germany. They were a great opportunity to meet people, connect with professors, and even make friends. You got familiar with what the other groups were working on while building nationwide connections.

Most importantly, compared to my friends at other institutions, we had less teaching responsibilities. I found that genuinely amazing — we could concentrate on research completely, which is exactly what you need during your PhD. That was a huge factor for me when I was choosing MPI-INF.

What advice would you give to students starting out in computer science, machine learning or computer vision, and to prospective SIC students – especially young women?

First things first, you need to build up your portfolio. When I first started as a HiWi, I was pushing to take on projects or help with existing ones; that’s how you learn. 

When you’re looking for a PhD, there are basically two options. First option: trying to work with a well-known, established professor, which can be great — but you have to understand you might not see them much because they’re busy. In this scenario, while it would also depend on the group structure, you’ll most likely be working quite independently, which can be challenging.

Your second option, which is very valuable but harder to find, is finding a fresh professor. They’re usually enthusiastic, more involved in your projects, and you can learn a lot from them. It can be slightly easier to get a position with a fresh professor, because there are fewer people applying to them. There are pros and cons to this though — firstly, as I said, it’s very hard to find. And you have to consider the institution and the resources it has. A fresh professor at a reputable institution is ideal. Fresh professors at less-established places, however… I’m not saying you can’t be lucky, but you could also be unlucky. It’s a hit or miss, you know? In a reputable institution, a fresh professor is more likely to give you attention and guidance. Senior professors, on the other hand, are less available because they often travel a lot and don’t attend all meetings — which is totally normal. Like I said, they’re busier in comparison.

If you feel too conflicted, asking peers for recommendations can help — although if they’re also master’s students, they might not know much themselves. All I’m saying is, it’s tempting to target the big names, I know, but it may not always be the ideal choice. Take your time and choose your destination wisely. I think I got really lucky with MPI-INF Saarbrücken. 

Another thing, and this is very important: don’t do ten projects at the same time — concentrate on one! One project is more important than ten simultaneously. It rarely works. I only know one person capable of doing that, and mind you, he’s a genius — but even he can only handle five at a time. Everyone else fails when trying to juggle more than one, myself included.

I was initially just a member, but now I’m on the board of the Women in Computer Vision workshops. We publish at major conferences like CVPR (Computer Vision and Pattern Recognition Conference), ICCV (International Conference on Computer Vision), and ECCV (European Conference on Computer Vision) rather than in journals.

From my experience, I would recommend women in computer science to simply reach out to other women — it’s a great way to meet people who can understand and support you. In our field, there’s a culture of guiding or helping out others when they’re at times of uncertainty — this applies to all genders, of course. But since women are still underrepresented in computer science, especially in computer vision, this support can be especially valuable.

As a matter of fact, many professors are trying to recruit more women for PhD positions, but it’s hard because there aren’t many applicants. So my advice is simple: just apply, that’s it! If you’re interested in someone’s research and would like to reach out, do it! Don’t be afraid to take the first step.

Interested in the Computer Vision and Machine Learning department of the Max Planck Institute of Informatics? Click here to visit their page.

Woman in Computer Vision? Visit the website of Women in Computer Vision to find out more about their activities!

Editor:
Saarland Informatics Campus Team
Email: yagmur.akarsu@uni-saarland.de

The following text has been machine translated from the German with no human editing.

Born in Eberbach am Neckar, he first completed an apprenticeship as a carpenter and joiner and went on to obtain an Ing. grad. in wood technology from the Rosenheim School of Engineering. He subsequently obtained his Abitur through adult education at the renowned Braunschweig-Kolleg and studied mathematics and physics at the University of Göttingen, as well as computer science at the University of Essex in the UK. There, he was awarded a Master of Science in Computing in 1972, obtained his doctorate in Artificial Intelligence in 1976 with a thesis on unification theory, and then returned to Germany.

Since then, Jörg Siekmann has worked first as a research assistant and, from 1980, as a university assistant at the Institute of Computer Science at the University of Karlsruhe, where he established a research group on artificial intelligence (AI) in the field of deduction systems and, together with others, founded the influential Collaborative Research Centre 314 on AI. In 1983, he was appointed to Germany’s first professorship in Artificial Intelligence at the University of Kaiserslautern, where, together with several colleagues, he founded the German Research Centre for Artificial Intelligence (DFKI). The Springer series ‘Lecture Notes in Artificial Intelligence’ (LNAI), which he founded, gained international prominence.

In the summer term of 1991, Siekmann accepted a post at Saarland University and, until 2006, held a dual role as Professor of Computer Science and as Head of Research at the German Research Centre for Artificial Intelligence (DFKI). On the Saarbrücken campus, his roles included serving as spokesperson for Collaborative Research Centre 378 ‘Resource-Adaptive Cognitive Processes’; in 1995, as a visiting professor at JiaoTong University in Shanghai, he initiated the cooperation with our university; together with his second wife and colleague Erica Melis, he served as scientific director of the ‘Centre for e-learning Technology’; and he was a senior professor and coordinator for digital education at Saarland University.

The honouree has made an outstanding contribution to establishing ‘Artificial Intelligence’ as a field of research in Germany and to creating the organisational framework for this discipline within the ‘German Informatics Society’, which awarded him the title of ‘Fellow’ in 2002 and later named him one of the ten most influential AI scientists. Overall, he has made significant contributions to AI research, taken on extensive responsibilities within the academic community, and advised research groups, governments and industrial companies at both national and international levels. His body of work comprises around 900 publications in his fields of research: deduction systems, proof planning, unification theory, e-learning for mathematics, as well as Buddhism and Artificial Intelligence.
Unfortunately, for health reasons, the honouree is currently unable to host public celebrations.

Further information: 
Dr Wolfgang Müller, University Archives
Email: wolfgang.mueller(at)uni-saarland.de, dr.wolfgang-mueller(at)t-online.de 
 

Presseeinladung zum Projektabschluss: 
To mark the conclusion of the Escade project, funded by the Federal Ministry of Research, Technology and Space, the team will present how AI can make a smaller environmental footprint while also giving small and medium-sized enterprises access to powerful AI models. The event will take place on 30 July from 10 a.m. to 1 p.m. at the DFKI on the Saarbrücken campus (D3 2) and is open to the public.

The following text has been machine translated from the German with no human editing.

Fast jeder nutzt heute auf die eine oder andere Weise Künstliche Intelligenz. Was nicht jedem bewusst ist: Auch die kleiAlmost everyone uses artificial intelligence in one way or another these days. What not everyone realises is that even the smallest response from a chatbot consumes energy and resources. Training and running AI models using vast amounts of data consumes hundreds of terawatt-hours worldwide. On a global scale, all of this leaves a massive ecological footprint. And with the rapid development of the technology, demand is set to rise sharply. New data centres will seal off vast areas of land, as will new power stations – including those fuelled by fossil fuels – which will have to supply the additional energy required. CO₂ emissions will rise just as dramatically as the water required for cooling.

A consortium led by Professor Wolfgang Maaß, who conducts research at Saarland University and the German Research Center for Artificial Intelligence e Artificial Intelligence (DFKI), is working to counteract this trend and make artificial intelligence more energy-efficient. For three years, the consortium has been investigating various methods and developing and testing new technologies. The researchers are presenting their findings at the conclusion of the project.

Compressed AI requires almost 90 per cent less energy

On the one hand, the team is focusing on smaller, more needs-based AI models to curb AI’s energy consumption and conserve resources. Today, AI uses huge data models. A chatbot, for example, utilises the entire data model – comprising trillions of parameters – to generate its response: figuratively speaking, it searches an entire library rather than just the books containing relevant content. The researchers have therefore developed AI models in which irrelevant parameters are not processed in the first place, making them more energy-efficient.

To do this, they filter out the knowledge that is truly necessary for the respective task from large teacher models and create bespoke student models that are up to 90 per cent smaller. “We’re achieving good results by compressing the AI models – in other words, making them smaller and more efficient. In our test runs, we were able to demonstrate that the student models deliver comparable performance whilst using up to 89 per cent less energy,” says Sabine Janzen, a postdoc in Wolfgang Maaß’s team. The leaner AI models, tailored to specific use cases, operate without the need for extensive infrastructure. “This makes powerful AI models accessible even to small and medium-sized enterprises, something that was previously impossible simply because of the size of the models,” says Janzen.

Automatically selecting the best AI model using 40 per cent less energy

For AI models that process and generate digital image data, the researchers use a different method known as ‘neural architecture search’. This method automatically identifies the best architecture for artificial neural networks. The team was able to demonstrate that they can reduce the size of the models by just under 90 per cent and cut energy consumption by 40 per cent. What’s more, the models do not suffer any loss of performance. “We were even able to improve the model’s accuracy with this approach,” says Sabine Janzen.

In machine learning using artificial neural networks, the learning processes are similar to those in the human brain. Whilst the human brain is a master of energy efficiency – having continuously optimised itself through evolution and processing information very efficiently – its artificial counterpart, with its efficient algorithms, requires an enormous amount of computing power and electricity: Artificial neural networks are still painstakingly assembled by humans today and fine-tuned until they deliver good results. “We are automating this process using neural architecture search. In doing so, we test various network structures and optimise them further so that the models are powerful and efficient, but cost less,” explains Sabine Janzen.

Scrap Sorting Test Case

In order to test such more energy-efficient AI methods on an industrial scale, the researchers have been collaborating with SHS – Stahl-Holding-Saar since 2022 as part of the Escade project.

SHS has developed a highly capable AI model that automatically classifies steel scrap and uses camera images to identify which type of steel scrap is being delivered to the steelworks site. Since 2024, SHS has been using its own in-house, optimised AI model to classify steel scrap: using camera images, the AI recognises the different types of scrap, enabling the steel scrap to be sorted by type before being used in production.

The aim of the Escade project was to make an AI model with roughly equivalent performance even more energy-efficient. The research team developed a high-performance visual AI model for steel scrap classification. This new model was compressed in such a way that it operates in a compact, energy-efficient manner whilst delivering performance similar to that of an AI model created using traditional methods. This makes the visual computing process more energy-efficient. To achieve this, the partners first trained the model using the complete data set and all relevant information, and then compressed it using knowledge distillation and automatically assembled neural networks. The techniques developed can, in particular, be applied to other visual AI models and can thus contribute to energy savings on a wider scale.

Further potential for savings in sustainable data centres

Furthermore, on 30 July, the consortium will highlight potential savings for data centres. Together with its partners, the Saarbrücken-based research team has developed a concept and recommendations for energy-efficient AI, which will enable data centres and AI users to plan more effectively and identify inefficient processes. To this end, the team has developed a tool that enables reliable forecasts of the exact energy consumption and costs of AI models. “This tool makes it possible, for example, to schedule processes that require high computing power for times when the price of electricity is low. Until now, decision-makers have found it difficult to estimate how much energy specific models will consume, which is why it is challenging to plan ‘ ’ in a cost-effective manner,” explains PhD student Hannah Stein, who is researching energy-saving AI methods.

Neuromorphic chip technologies

In addition, the research consortium is working on neuromorphic chip technologies in the field of hardware; these are microprocessors that also mimic the functioning of the human brain. “Our findings in this hardware area suggest that this technology also operates significantly more energy-efficiently than conventional chips: in our tests, they are already up to six times more efficient. But further research is needed here, and we need more time for that,” explains Sabine Janzen.

On 30 July, the research consortium will present its project findings on sustainable AI data centres and resource-efficient AI algorithms, including through demonstrations. Interested parties, in particular data centre operators, technology providers and companies, are invited to attend.

Registration and programme for 30 July: https://escade-project.de

Background

The ESCADE (Energy-Efficient Large-Scale Artificial Intelligence for Sustainable Data Centres) project was funded by the Federal Ministry of Research, Technology and Space (BMFTR) with around five million euros over a three-year period as part of the ‘Green Tech Innovation Competition’.

The project consortium comprises, in addition to the research team led by Professor Wolfgang Maaß as coordinator (German Research Center for Artificial Intelligence DFKI and Saarland University), the Technical University of Dresden, Bielefeld University, the Central German Data Centre NT Neue Technologie AG (NT.AG), SHS – Stahl-Holding-Saar GmbH & Co. KGaA, SEITEC GmbH, the Austrian research organisation Salzburg Research, and the subcontractors eco2050 Institute for Sustainability, SpiNNcloud Systems GmbH and elevait GmbH & Co. KG.

The project is managed by the German Aerospace Centre.

For enquiries, please contact:

Dr Sabine Janzen: Tel: 0681- 8 57 75 – 269, Email: sabine.janzen@dfki.de

Hannah Stein: Tel: +49 681 302-64739, Email: hannah.stein@iss.uni-saarland.de

https://escade-project.de

Every year, millions of deaths worldwide are linked to antibiotic resistance. At the same time, developing new antibiotics is particularly risky, as many drug candidates are eliminated in early development phases due to undesirable side effects or lack of efficacy.

OptimAIze addresses the urgent medical need for new antibiotics and the specific development challenges posed by antimicrobial resistance and high preclinical failure rates, often caused by an unfavorable balance between efficacy and toxicity. The goal is to develop and apply generative AI methods to target the modification of drug candidates to increase their efficacy and reduce cytotoxicity—that is, the potential damage to human cells. A particularly innovative aspect is the integration of mechanistic knowledge into the AI.

“Artificial intelligence opens up enormous opportunities for drug discovery. At the same time, purely data-driven models reach their limits, especially when only limited or unbalanced data is available. That is why, in OptimAIze, we rely on a combination of AI and mechanistic expertise. This allows us not only to make more precise predictions but also to better understand why certain molecules are effective or toxic.”

Prof. Dr. Verena Wolf, Head of Neuro-Mechanistic Modeling, Project Coordinator OptimAIze

At the heart of the project is a novel closed-loop learning cycle that combines artificial intelligence and experimental validation. High-resolution cytotoxicity data are analyzed using modern AI methods. In the process, the models not only learn to predict the efficacy and tolerability of molecules but can also specifically suggest new drug candidates, which are then tested experimentally. The results are fed back into the AI models, continuously improving their predictive power.

The added value of OptimAIze lies in its integrated AI optimization, which combines knowledge across AI and computational chemistry to efficiently propose molecules with improved antibiotic activity and lower toxicity that can be synthesized and tested immediately.

To this end, the consortium brings together complementary expertise: DFKI develops methods for explainable and generative AI; Saarland University contributes expertise in drug design, language models, and single-cell analyses; HIPS handles the chemical synthesis and optimization of the candidates; and smartbax GmbH provides exclusive data and drug programs from industrial antibiotic research. 

In addition to developing specific antibiotic candidates, OptimAIze also aims to make innovative AI methods and software tools available to the scientific community. The algorithms developed will be published as open source—as far as possible—and will thus support further research projects in the long term in the fight against antibiotic resistance.

With a duration of three years, OptimAIze aims to set new standards for AI-supported drug discovery and make a significant contribution to combating antimicrobial resistance. The project is funded by the Federal Ministry of Research, Technology, and Space under the funding guideline “Application of Artificial Intelligence in Drug Discovery.”

The university’s project, Saarland Future Minds Programme in Quantum Technologies and Artificial Intelligence, was selected from more than 70 applications in a nationwide competition organized by the German Academic Exchange Service (DAAD). The DAAD is funding the project with around 709,000 euros through the end of 2029.

The following text has been machine translated from the German with no human editing.

With the recently launched Centre for Quantum Technologies, the state and Saarland University have underlined their aim of establishing this key future technology in the Saarland – not only within the research landscape but also beyond it, for example in spin-off companies. The German Academic Exchange Service (DAAD) is now supporting this initiative with comprehensive funding to attract students and doctoral candidates, who are to be specifically recruited and retained.

The “Saarland Future Minds Programme in Quantum Technologies and Artificial Intelligence” (FMP Quantum AI) highlights Saarland University’s status as a leading national centre for attracting outstanding young talent to future technologies. “At the same time, our success in the ‘Academic Horizons’ funding scheme is a testament to Saarland University’s successful international focus on key technologies,” emphasises University President Professor Ludger Santen.

Under the project leadership of Professor Moritz Weber, the scientific director of the Centre for Quantum Technologies, the DAAD funding will be used to establish structures that will enable three key objectives: Firstly, the aim is to identify the best Master’s students and PhD candidates internationally in the fields of quantum technologies and artificial intelligence. Key tools in this regard are international partnerships and the establishment of international summer schools with research placements at the university. Secondly, these talented individuals will be supported with comprehensive assistance from initial contact right through to the start of their studies, in order to systematically promote their academic and social integration. And thirdly, these outstanding young researchers will be supported in their career development and retained in Germany as a research hub in the long term.

Jakob von Weizsäcker, Minister for Finance and Science of the Saarland, comments: “Advances in artificial intelligence and quantum technologies are driven by exceptionally talented and committed individuals. Attracting these talents to the Saarland is a key factor in our location’s appeal. The fact that Saarland University is among the universities selected nationwide is a mark of distinction for our research hub and underlines its international appeal.”

The Future Minds Programme Quantum AI thus makes a direct contribution to the DAAD’s ‘Academic Horizons’ programme. The aim of this programme is to identify and integrate outstanding international talent, to support their academic careers, and to strengthen the German innovation landscape in key technologies of the future – in this case, quantum technologies and AI – in the long term. The DAAD is funding the Quantum AI project from 1 May 2026 until the end of 2029 with a total of 709,000 euros.

Further information on the DAAD’s Academic Horizons programme:  https://www.daad.de/de/infos-services-fuer-hochschulen/weiterfuehrende-infos-zu-daad-foerderprogrammen/academichorizons/

For enquiries, please contact:
Prof. Dr Moritz Weber
Tel.: 0681 302-2556 
weber(at)math.uni-sb.de

From June 24 to 26, 2026, Saarbrücken was once again a hub for young computer science enthusiasts from all over Germany. Since 2008, the Saarland Informatics Campus, as the host of the “Saarbrücker Forschungstage”, has been offering selected high school students the opportunity to delve deeper into the world of computer science. The event is aimed at 14- to 20-year-olds who have distinguished themselves through exceptional achievements in the subject and are seeking an introduction to academic computer science. 

The goal of the Research Days is to inspire young talent to pursue a degree in computer science, to provide them with early exposure to research, and to support them on their path into academia.

This year, 42 young Germans participated, including 6 young women, as well as two students from a German school in Romania. As always, the invitees include the top participants from the second round of the 44th National Computer Science Competition (Bundeswettbewerb Informatik), as well as outstanding high school graduates who have particularly impressed in their advanced computer science and mathematics course.

The first meeting on the afternoon of June 24 was followed by a pre-event evening to set the tone and provide an opportunity for participants to get to know one another. This was followed by two days of exciting presentations, lectures, and workshops, each running from morning until late afternoon. On Thursday, an evening event titled “Ethics for Nerds” rounded out the research days thematically. The students could choose from 6 broader topics to create their own program; each participant attended three different workshops in small groups. The topics covered a wide range: from theoretical computer science and satellite communications to bioinformatics, game programming, and quantum computing as well as a session on AI-large language models. 

This year, the workshop instructors come from the Department of Computer Science of Saarland University, the Max Planck Institute for Informatics, the Max Planck Institute for Software Systems, the Center for Bioinformatics, and the CISPA Helmholtz Center for Information Security.

Participants also attended an advanced lecture program featuring presentations by Professors Thorsten Herfet (Saarland University) and Joël Ouaknine (MPI for Software Systems), who provided insights into their current research.

The Saarbrücken Computer Science Research Days are organized in close cooperation with the Nationwide Computer Science Competitions (Bundesweite Informatikwettbewerbe) and pursue the common goal of inspiring young people interested in computer science to engage with the subject. The German National Computer Science Competition has taken place annually since 1980 and is hosted by the German Informatics Society, the Fraunhofer ICT Group, and the Max Planck Institute for Informatics. The competition is supported by the Federal Ministry for Education, Family Affairs, Senior Citizens, Women and Youth.

Further Information:
Website of the Nationwide Computer Science Competitions: https://bwinf.de/

Editor:
Bertram Somieski
Tel.: +49.681.9302-5710
Email: somieski(at)mpi-klsb.mpg.de

The project addresses a fundamental challenge in modern cyber-physical systems and has far-reaching implications across a wide range of applications such as autonomous driving, robotics and space exploration, where safety and reliability are paramount.

The ERC-funded research will go beyond verifying that the theoretical design of control systems meets its intended specifications, such as driving at a certain speed or maintaining a certain distance from other vehicles. It will also ensure that these guarantees remain valid for the implementation of the systems, despite the timing variations that inevitably occur during execution.

Self-driving cars, increasingly sophisticated robots, and autonomous spacecraft such as autonomous spacecraft and planetary exploration vehicles are all examples of highly complex systems governed by advanced computer control. Their growing capabilities depend on ever more finely coordinated computations and increasingly precise timing. Yet timing remains a critical source of vulnerability. Even minor deviations in the synchronization of computational processes can propagate through a system and, in safety-critical applications, lead to serious or even catastrophic failures.

“Even the most advanced computer-controlled systems can be vulnerable to timing-related faults,” says Martina Maggio. “The Mars helicopter Ingenuity, for example, experienced dangerous oscillations caused by timestamp misalignments and later encountered another synchronization issue. These incidents demonstrate that timing errors are not rare edge cases but remain a fundamental challenge in the design and operation of complex autonomous systems.”

Despite these setbacks, Ingenuity ultimately became a remarkable success. During one incident, the helicopter was forced to perform an emergency landing after an image-processing task fell out of sync with other onboard computations. Maggio’s research focuses on preventing precisely such failures in software-controlled technical systems, from autonomous vehicles and industrial robots to spacecraft. This is particularly important in applications where correct timing is as important as correct functionality: if the timing of critical events cannot be guaranteed, the consequences can range from degraded performance to catastrophic system failure.

To advance this research at the highest international level, Martina Maggio has been awarded an Advanced Grant from the European Research Council (ERC). Since joining Saarland University as a Professor of Computer Science in 2020, Maggio has established herself as a leading expert at the intersection of control engineering and real-time systems. The ERC-funded project, entitled SCARF (Scalable CPS Analysis of Robustness to Failures), will begin in 2027 and run for five years.

Further information:

Prof. Dr Martina Maggio 

Email: maggio(at)cs.uni-saarland.de

Whether it is online banking, email, video streaming, global cloud platforms, or large-scale AI infrastructures – planetary-scale distributed systems form the backbone of many societal-scale applications. We take for granted the continuous availability of these services, even though outages can cause widespread disruption. Yet today, developers do not have principled approaches to provision, analyze, or prove performance or resilience properties of such systems. Currently, developers test for such properties using expensive but inadequate workload testing, and availability outages continue to be a problem for users.

The now EU-funded project “Pascal: Formal Performance Analysis at Scale” addresses the major challenge of formally reasoning about (that is, mathematically describing and verifying) the performance and resilience of large-scale distributed systems. Its ultimate goal is to develop methodologies and tools that system developers can use to reason about implementations of their systems.

Rupak Majumdar has been a Scientific Director at the Max Planck Institute for Software Systems since 2010 and an Honorary Professor in the Department of Computer Science at the Rhineland-Palatinate University of Technology Kaiserslautern-Landau (RPTU). This is the second ERC grant he has received. In 2015, together with Michael Backes, Peter Druschel, and Gerhard Weikum, he was awarded an ERC Synergy Grant for the project ImPACT: Privacy, Accountability, Compliance, and Trust in Tomorrow’s Internet. The ERC Synergy Grant is the European Research Council’s most highly funded grant scheme.

More info from the ERC.

Lenssen, who is currently a researcher at the Max Planck Institute for Informatics on the Saarbrücken campus, will further strengthen the Saarland Informatics Campus.

The ever-changing nature of the real world poses a challenge for artificial intelligence. Today’s AI systems are particularly well suited to text: they cope well with unambiguous, so-called ‘discrete’ data – that is, separate values that are countable and cannot grow continuously – as well as with symbolic information. In the real world, however, many things get out of their control. Here, much of the information is in continuous form: within a certain range, the values change constantly, such as sensor data or the movements of objects and people.

In his research, Jan Eric Lenssen explores how artificial intelligence can utilise and generate data that is present in so-called continuous representations – such as video, time series or three-dimensional sensor data. The Computer Science researcher develops methods of visual artificial intelligence that enable neural networks to process, analyse, model and generate such complex structures – a task that is significant for fields including robotics, physical and generative artificial intelligence, and image processing.

Jan Eric Lenssen became known, amongst other things, for PyTorch Geometric (PyG), a software library for graph-based neural networks, which he co-developed during his PhD at TU Dortmund University. PyG is now the world’s most widely used library of its kind and forms the technical basis for numerous research projects and practical applications. The start-up kumo.ai, based in Mountain View, California, which specialises in machine learning for relational databases and of which Lenssen was a founding member, also built upon this library.

At Saarland University, Lenssen is contributing his expertise in computer vision and artificial intelligence and aims to further expand his internationally successful research profile at the Saarland Informatics Campus. Through the joint research centre of the Max Planck Institute and Google, as well as a long-standing cooperation programme with Intel and the Saarland, his work is closely integrated with the international research landscape and industry.

Jan Eric Lenssen has received numerous awards for his research, including the DAGM German Pattern Recognition Award in 2025 and an Emmy Noether Fellowship worth around 1.9 million euros. His scientific work has also received high-profile accolades at leading conferences on computer vision and machine learning, including a Best Paper Award at ECCV 2022 and a Best Paper nomination at CVPR 2020. For his [doctoral] dissertation, he received the ECVA PhD Award and the TU Dortmund University Dissertation Prize.

Short biography 

Jan Eric Lenssen studied Computer Science at TU Dortmund University from 2009 to 2015, where he graduated with distinction. He subsequently obtained his doctoral degree there in 2022 with the highest distinction, ‘summa cum laude’. He undertook research stays at Meta Reality Labs (USA) and the AI company nnaisense (Switzerland), and from 2021 onwards was part of the founding team of the start-up kumo.ai. He then conducted research as a postdoctoral researcher and later as a senior researcher and research group leader at the Max Planck Institute for Informatics in Saarbrücken. Since 2026, Jan Eric Lenssen has been Professor of Computer Science at Saarland University and an associate member of the Max Planck Institute for Informatics.

For enquiries, please contact:

Prof. Dr Jan Eric Lenssen, email: jlenssen@mpi-inf.mpg.de 

Press photos available for download: 

Press photos can be found on this news webpage: https://www.uni-saarland.de/universitaet/aktuell/news.html

Derek Dreyer has set an exemplary standard of service within the PL community. Throughout the years, Derek has served in almost every possible role in our community, including General Chair of ICFP’19, Program Chair of POPL’24, and Associate Chair of OOPSLA’23 and PLDI’26. Since 2017, he is an Associate Editor of TOPLAS; since 2023, he is on the PACMPL advisory board; and since 2022, he is co-editor-in-chief of the Journal of Functional Programming, presently overseeing the transfer of the journal to Diamond Open Access. But Derek’s service work began much earlier. While still an assistant professor, he was moderator of the TYPES mailing list and volunteered to serve on the SIGPLAN Executive Committee (2012-15) as Awards Chair. In 2020, he helped safeguard the perception of programming languages research outside the field, when the CORE ranking committee threatened to demote the rankings of several major PL conferences. Derek has also been a prominent mentor for students, postdocs, and junior faculty. He chaired early editions of PLMW and co-founded the RTFM workshop on faculty mentoring (at PLDI’24 and POPL’26). He frequently speaks at such events, and is known for his widely-cited talks on speaking and writing skills: “How to Write Papers and Give Talks That People Can Follow”. Last but not least, Derek has written popular blog posts touching on the more “human” aspects of a career in research such as: accepting criticism, handling rejection, struggling to maintain a work-life balance, and impostor syndrome. In summary, Derek has served the community in many ways, always striving to promote the value and the quality of PL research, to help maintain cohesion in our community, and to empower junior researchers to conduct outstanding PL research.

The Distinguished Service Award is given by ACM SIGPLAN to recognize distinguished service contributions to the Programming Languages Community. The award recognizes contributions to ACM SIGPLAN, its conferences, publications, or its local activities. The award includes a prize of $2,500.

The report by the United Nations University (UNU) examines the environmental impacts of artificial intelligence, focusing primarily on generative AI models such as ChatGPT, Claude, and DeepSeek. Among other things, the study examines the resource requirements of data centers for training and operation, as well as the associated CO₂ emissions. As the use of generative AI increases, so does the computing capacity required for inference and training.

Prof. Maaß generally welcomes the debate but is critical of parts of the report. “The report makes a useful contribution by systematically compiling consumption figures for the first time and making them accessible to a broader audience,” he says. At the same time, he emphasizes: “The figures cited in the report are plausible in terms of their magnitude, but difficult to reproduce methodologically.”

Prof. Maaß generally welcomes the debate but is critical of parts of the report. “The report makes a useful contribution by systematically compiling consumption figures for the first time and making them accessible to a broader audience,” he says. At the same time, he emphasizes: “The figures cited in the report are plausible in terms of their magnitude, but difficult to reproduce methodologically.”

This is precisely where the ESCADE project, funded by the Federal Ministry for Economic Affairs and Climate Action, comes in. As part of this project, DFKI is researching energy- and cost-efficient approaches for operating AI applications in data centers. Using the EAVE demonstrator, energy consumption, CO₂ emissions, and operating costs of different model, hardware, and location configurations are made comparable.

In doing so, ESCADE addresses key transparency and decision-making issues that are also relevant in the context of the UN report. The scientific foundations and design principles were described, among other places, in a CAiSE publication on energy- and cost-efficient AI operations. In this way, the project helps not only to discuss the environmental impacts of AI but also to make them measurable and practically comparable.

“The energy consumption of AI data centers is real and growing, but it is not currently the dominant climate issue,” says Maaß. It is therefore crucial to consider technical development, site selection, and regulatory frameworks collectively. Against this backdrop, the question of how AI systems can be designed in the future so that their benefits do not come at the expense of unnecessarily high resource consumption is gaining importance. 

The UNU-INWEH report:https://unu.edu/inweh/collection/environmental-cost-of-AIs-Enrgy-Use-Carbon-water-and-land-footprints

Statements on the UN report:https://www.sciencemediacenter.de/angebote/un-bericht-zum-umweltfussabdruck-von-ki-und-rechenzentren-26129

Paper „Towards Decision Support Systems for Cost-Effective and Energy-Efficient AI Operations in Data Centers“: https://link.springer.com/chapter/10.1007/978-3-032-28110-4_22 

ESCADE project webpage: https://www.dfki.de/web/forschung/projekte-publikationen/projekt/escade