Conference Agenda

Research data infrastructures have become increasingly important for mathematics. MIMIR (Multi-Indexed Mathematics Information Retrieval) is an ongoing project at the Institute of Mathematics of the University of Potsdam that investigates how mathematical knowledge can be represented as a structured data lake. The central idea is to preserve provenance, multiple representations, and source-specific information while enabling the gradual construction of semantic relationships. The project is developed together with master's students and serves both research and educational purposes. I will present the vision, architecture, and current status of the project, as well as its connections to mathematical knowledge graphs and future AI-supported systems.

Formal query syntax is a powerful but often intimidating way to explore zbMATH Open data. Since June 2026 the service has been launched as a live, switchable option in the one‑line search field: users can type ordinary sentences such as “Which arxiv preprints from the last 5 years deal with the ABC conjecture?” and the system automatically translates the input into a Boolean combination of facet queries (MSC codes, author names, arXiv tags, etc.). The translation is performed by a large‑language model that runs on the GWDG KISSKI LLM service. The tool is not a conversational chatbot - it processes a single query, returns the corresponding hits, and then stops. In the talk we will (i) demonstrate the interface and the feedback mechanism that lets users report queries that did not yield satisfactory results, and (ii) discuss the current limitations (coverage of only document‑level facets, rate limits, and occasional variability of the model's output). Attendees are invited to try the feature and contribute feedback to shape the next generation of AI‑enhanced mathematical search.

Mathematical texts, often learning texts, are generally “perceived” graphically: a receiver reads the texts and formulas with his or her eyes and makes sense of it. How do persons using a different channel do?
The current technology makes it possible to claim to “just read-aloud” but a lot of hurdles remain before the perception process is on par with the reading process when mathematics is involved. Formulas in particular can be read-aloud with specialised systems but we are far from a mainstream practice in many languages.
The case that interests us is the learning process in a university of applied sciences where many different mathematics is being taught. The presentation will report on a survey on the expectations and the experience so far. We aim at answering questions such as:

• How complex can a formula be, for it be understandable when first read?
• How do various tools compare in their fidelity, trust, and understandability when reading which formulas?
  In particular, can generative AI bring enhancements or losses?
• Which usage situation is plausible to listen to texts with formulas? Jogging? Learning at a desk without reading? Driving?

The presentation will demonstrate how MathML-enabled read-aloud works with state of the art technology and what enhancements can be made to make the perception of a text with mathematics an integral part of an effective learning process. In particular, it will echo scientific experiments which indicate features or aspects in very diverging manners.