Final Conference

Interpretations, Explanations, and Knowledge Gain

March 11 – March 12, 2024

The semester is organized within the framework of the Berlin Mathematics Research Center MATH+ and supported by the Einstein Foundation Berlin. We are committed to fostering an atmosphere of respect, collegiality, and sensitivity. Please read our MATH+ Collegiality Statement.



The presentation of the results of a mathematical analysis can often be done in different ways. Since “small data” leads or should lead to decisions/actions, the presentation plays an important role. Small data sets may not contain “a lot of data”, but they do contain “many possible interpretations”. The final conference covers the aspects of “interpretation, explanation and knowledge gain”. Mathematical, methodical representations are sought that depict the wealth of possible interpretations. Here again (as within the previous semester activities), an interface or cooperation between different disciplines must be constructed.

Program (Preliminary)


Monday 11.3.2024

Venue: Zuse Institute Berlin, Takustraße 7, 14195 Berlin


09:00 – 10:00    Welcome / Onsite Registration

10:00 – 10:30    Introduction – Marcus Weber (ZIB)

10:30 – 11:15    Decision Theatre – Sarah Wolf (FU Berlin)

11:15 – 12:00    Learning from Small Data by Patch Normalizing Flow Regularization – Fabian Altekrüger (HU Berlin)

12:00 – 13:30    Lunch Break

13:30 – 14:15    Small Data and AI for Exploring Chemical Space  – Christopher Secker, Konstantin Fackeldey (TU Berlin)

14:15 – 15:00    MaRDI, NFDI, and knowledge graphs – Karsten Tabelow & Thomas Kubrucki (WIAS), Marco Reidelbach (ZIB)

15:00 – 15:30    Coffee Break

15:30 – 16:15    Geometric learning for quantitative analysis of stone tool reduction sequences – Christoph von Tycowicz, Elodie Maignant, & Julius Mayer (ZIB)  

16:15 – 17:00    Get-Together


Tuesday 12.3.2024

Venue: Villa Engler, Altensteinstein 2, 14195 Berlin


09:00 – 10:30    3 Parallel Sessions

        • Small Data Analysis – Marcus Weber
        • Decision Theatre in action – Sarah Wolf
        • Learning from Small Data – Fabian Altekrüger

10:30 – 12:00    3 Parallel Sessions

        • Exploring Chemical Space with AI
        • Mathematical Research Data Initiative: Vision and Services– Marco Reidelbach
        • Geometric Learning – Christoph von Tycowicz

12:00 – 13:00    Lunch Break

13:00 – 14:00    Keynote – David A. Smith (Northeastern U) and Ryan Charles Cordell (U Illinois)

All Datasets are Small (if You Zoom Far Enough)

In the 19th century US, newspapers across the country and the world were linked through the “exchange system,” through which editors swapped papers for mutual reuse. The Viral Texts project ( has studied this system through many lenses, but this talk will focus on a few central questions of scale and knowledge representation. Given the losses of history and biases of digitization, how can we contextualize our findings within the larger exchange system we know existed? In the digital humanities there has long been a discourse advocating “scalable” or “zoomable” reading that negotiates between scales of analysis, but how does this work in practice? How should analyses at scale—e.g. classification models for reprinted genres, network graphs of information flow, or aggregated data about formal trends in newspaper publication—inform closer readings of specific reprinted texts, genres, editors, or newspapers? We present models of the textual composition of newspapers, the prevalence of exchanges, the periodic circulation of information, and the spatial reach of different texts to provide a view of these broad structures and the individual editorial decisions that produced them. The overlapping of telegraphic urgency, literary judgment, and political partisanship with local and international concerns makes the nineteenth century newspaper a chimerical model organism with implications for earlier periods, and ideas about manuscript culture, to our present concerns about social media and mechanically multiplied patchworks of content.


14:00 – 14:30    Keynote – Youssef Nader (FU Berlin)


Resurrecting Ancient Scrolls with AI

Recent advances in X-ray tomography and computer vision have enabled us to virtually unwrap ancient scrolls and look inside of them. The Vesuvius Challenge is an open challenge to leverage these advances to read Herculaneum papyri that was buried under volcanic mud.
In this talk, I will present my solution to using AI to read letters from within the scrolls which was awarded a first letters prize by the Vesuvius Challenge. I will talk about the data and challenges of reading carbon-based ink as well as how I was able to scale up the model and data to read more than two thousand letters from inside of the scroll and win the Grand Prize.

14:30 – 15:30    Fish-Bowl-Discussion

15:30 – 17:00    “Goodbye Thematic Einstein Semester“



Goodbye “TES Small Data”

The Thematic Einstein Semester on Small Data Analysis ended with a Final Conference about “Interpretation, Explanation, and Knowledge Gain”.  Mainly mathematicians from Berlin and from all MATH+ institutions (TU, HU, FU, WIAS, ZIB) summarized and presented their own point of view regarding possible approaches towards small data analysis (first day at ZIB). We focused on knowledge gain and on how to explain and represent possible interpretations of data.  It has been an “offline” conference. In total 37 participants took the chance to intensively discuss various key topics in small groups (second day at Villa Engler). The last three keynotes from international speakers addressed the question about benefits from interdisciplinary cooperation in data analysis.