AI to accelerate Scientific Understanding

Machine learning models have demonstrated a vast capacity to learn complex phenomena from data and to reproduce them accurately, for example by providing precise predictions of molecular properties or by generating human language in natural conversations. However, due to their black-box nature, such models often resemble oracles with vast stores of knowledge rather than scientific tools capable of reporting or communicating the reasoning underlying their predictions. This limitation constrains their potential impact in scientific applications, where understanding why a model arrives at a prediction can be as important as the prediction itself and can substantially accelerate scientific discovery and understanding.

This workshop seeks to develop a broader and more systematic understanding of how the explanation and interpretation of artificial intelligence models can enable AI-driven research from both theoretical and applied perspectives. In particular, the aim is to explore how AI-driven research can be enriched beyond model validation and trust assessment. The workshop will include invited talks, poster sessions, and dedicated time for discussion, fostering informal exchange across disciplines such as molecular science, medicine, digital humanities, geoscience, and related fields.

The aim of the three-day-workshop plus an optional preceding half-day tutorial is to bring together internationally renowned scientists, local scholars, and students. The tutorial offers a concise introduction to paradigms and tools that enrich AI-driven research with transparency and understanding. Through hands-on discussions, practical software demonstrations, and concrete application examples, participants will gain familiarity with key explanation methods for machine learning models. No prior experience is required – the tutorial is designed to ensure everyone can follow the main workshop with confidence.

Speakers include Dagmar Kainmüller, Helmholtz Imaging’s Spokeperson and Head of Research Unit at MDC.

More information including speakers & registration

The workshop is jointly organized by ELLIS Unit Berlin and BIFOLD – The Berlin Institute for the Foundations of Learning and Data.