Program
13:00 – 13:05
Welcome
Joachim Giesen
Friedrich Schiller University Jena
13:05 – 14:00
Keynote: The Modern Era of Deep Learning: From Foundations to the Present
Sören Laue
University of Hamburg
Abstract: Over the past fifteen years, deep learning has evolved from early image classifiers to powerful language models and agentic systems capable of multi-step reasoning and tool use. This talk revisits the key turning points along that trajectory: the rise of GPU-accelerated training, convolutional and residual networks, the transformer architecture, diffusion models, scaling laws, and scientific milestones such as AlphaFold. Rather than speculate about the future, the presentation analyzes how shifts in data, compute, and architectural design reshaped both capability and perspective. As scaling approaches practical and conceptual limits, a deeper understanding of learning dynamics and architectural behavior becomes increasingly essential.
14:00 – 14:45
Company Exhibition
14:45 – 15:10
Diversifying the R Ecosystem -- New Approaches for Package Binaries & Content Publishing
Patrick Schratz
devXY, Switzerland
Abstract: Patrick’s talk explores the motivation behind rpkgs.com, a new open-source approach for building arch-agnostic R package binaries on Linux (including Alpine!). He also demonstrates how these are integrated into ricochet (ricochet.rs), a new app for publishing static and dynamic content for R, Python, and Julia.
15:10 – 15:35
Data-driven Digital Evolution of Modern Laboratories
Marta Dembska
German Aerospace Center
15:35 – 16:00
Applied Data Science in Software Consulting: Trench Detection and Location Clustering
Sebastian Wuttke & Paul Kahlmeyer
TNG Technology Consulting
Abstract: Software consulting presents unique data science challenges where problems often lack the precise formulations common in academic settings. This talk presents two case studies from TNG that demonstrate how practical solutions emerge through domain-driven problem structuring. The first case addresses trench path extraction from large-scale LIDAR point cloud data for an infrastructure project. Without an explicit mathematical model for trench detection, we rely on geometric assumptions like elevation contrast, local density, linearity, and global tree topology. These are translated into a robust processing pipeline that operates within strict runtime constraints, illustrating how imposing structure on an ill-defined problem yields tractable solutions. The second case tackles venue optimization for a distributed workforce. We examine Affinity Propagation, the metric k-Median problem, and a strategic search space reduction approach, demonstrating how balancing theoretical models with practical feasibility enables real-world deployment. Together, these cases highlight a core insight from applied data science: effective solutions often emerge not from optimizing predefined objectives, but from carefully introducing domain-specific structure to complex, ambiguous problems.
16:00 – 16:45
Company Exhibition
16:45 – 17:30
Capstone (TBA)
Thomas Wolfers
Friedrich Schiller University Jena / University of Tübingen
17:30 – 19:30
Get-together
Snacks and drinks
