In this seminar, our SmallData Associated Researchers will give a 30 minute presentation of their work on small data and explore how it connects with other projects across the Collaborative Research Center. Each seminar will feature an open Q&A session.

The Tandem Lecture Series, hosted by SMART, highlights collaborative research performed by the Prinicpal Investigators across different disciplines. In each session, PIs will present their joint project, discuss developments since the project began, and share their disciplinary approaches. They will also reflect on their experience of working together, including what has helped and what challenges they faced in building a shared understanding. The sessions will conclude with a Q&A, giving researchers, both early career and beyond, the chance to ask questions and join the conversation.

In this collaborative lecture, Rolf Backofen and Toni Cathomen of Project A05 will be presenting.

SMART Doctoral Researchers, Early Career Researchers (ECRs), and Associated ECRs will be visiting the Sorbonne Uni in Paris, involving the Sorbonne Center for AI, LIP6 Computer Science lab and Laboratory of Computational and Quantitative Biology. We have an exciting program with enriching talks and networking opportunities jointly with our hosts.

This working group has developed out of the SmallData Seminar Series: Overview Paper Seminar on model-based similarity and its quantification with small data, moderated by Nadine Binder and Johannes Hertel.

Moderators: Nadine Binder (A02) & Johannes Hertel (B06)

In this workshop we approach uncertainty in various task from the exploratory viewpoint – on the one side we want to gather problems and questions in the small data setting with inherent uncertainty, offering you the possibility to show your questions and research. And on the other side we want to present already existing approaches and methodologies to help you solving your problems and enhancing existing approaches.

Moderator: Thorsten Schmidt (A03)

Large language models (LLM) can deal with information which for long time has been thought to only be processable by humans with a given contextual knowledge. Being neural networks, LLMs learn to quantify context sensitive similarity of textual information, also referred to as attention. Specifically by employing this similarity mechanism on multiple layers of abstraction, LLMs allow to address a certain level of a lack in structure and heterogeneity, e.g., in writing style and employed nomenclature, because similarity can be computed on the level of the semantics.

Still, a lack of structure and heterogeneity poses problems when small amounts of information are hidden in a large text, i.e., a low signal to noise ratio, which is the case for scientific literature on rare diseases or when a large heterogeneity is observed such as in doctors’ reports.

In this workshop we want to identify/develop specific use cases which represent the above described challenges from SmallData and develop approaches which (1) address the challenges and (2) quantify the uncertainty of the extracted information.

We give clinical researchers the opportunity to present their challenge of extracting information from heterogeneous text sources. These can be scientific publications such case reports or doctors’ reports describing the state of individual patients.

Participants from biomedicine describe the data, the specific problem and indicate the hurdles with respect to data protection. In addition they also indicate a panel of experts which is willing to validate the results retrieved from the LLMs.

Participants from computer science and statistical data science then infer feasible solutions for the challenge and develop a methodological research question that is in line with the SmallData objectives.

Goal of the workshop is to establish availability of datasets, specific biomedical use cases and strategies for validation of extracted information.

Moderator: Fabian Kabus (F)

The Tandem Lecture Series, hosted by SMART, highlights collaborative research performed by the Prinicpal Investigators across different disciplines. In each session, PIs will present their joint project, discuss developments since the project began, and share their disciplinary approaches. They will also reflect on their experience of working together, including what has helped and what challenges they faced in building a shared understanding. The sessions will conclude with a Q&A, giving researchers, both early career and beyond, the chance to ask questions and join the conversation.

The speakers for this collaborative lecture will be announced shortly.

On July 18, we’re hosting the SmallData No Rules Science Exchange focused on networking with the opportunity to check out the SmallData Companion and Cubie, our new chatbot. We’ll kick off with a Companion and Cubie Meet & Greet, followed by informal exchange. This is a space for all SmallData members to connect and share ideas for the next funding period.

This workshop discusses different application settings and concepts of synthetic data in a small data context. We will focus on data types such as temporal and longitudinal data, data with a hierarchical structure, and disease gene prioritization. In addition to discussing whether synthetic data underestimates uncertainty and how to use synthetic experts to assess the quality of generated data, we will demonstrate the models and tools currently in use to generate these data types.

Moderator: Clemens Schächter (A03)

On 4 August, 2025, participants of the SMART Master Lab Cohort 2.0 will gather at the University of Freiburg for the kick-off event. This event offers an opportunity to meet, network, and build connections with fellow SmallData researchers from diverse fields, including biomedical sciences, computer sciences, mathematics, and systems modeling. It’s a chance to lay the foundation for future collaborations, foster interdisciplinary relationships, and kick-start the exciting work ahead in the SMART Master Lab!