Title: On the Use of Foundation Models for Microscopy and Structured Medical Data Analysis

Abstract:

Foundation models provide a general framework for learning transferable representations from large datasets and their use for medical data offers large opportunities. Medical domains pose distinct challenges, including limited labeled data, domain shifts, and heterogeneous data modalities. I will present our research in Berlin, which investigates the use of foundation-model approaches in two complementary settings: microscopy image analysis and structured medical data. In the imaging domain, we study how synthetic data can support the training and adaptation of models, particularly in scenarios with limited annotations. In parallel, we examine representation learning for structured medical data. Across both modalities, we analyze the extent to which foundation models enable transferable representations and robust generalization, and we discuss implications for interpretability and data efficiency. The results highlight both the potential and the current limitations of foundation-model approaches in medical data analysis, and point to open challenges.

Bio of the speaker:

Erik Rodner is a professor of Machine Learning and Data Science at the University of Applied Sciences Berlin (HTW Berlin). His work lies at the intersection of machine learning and computer vision, with a particular interest in learning under limited data availability. Prior to joining HTW Berlin, he held a tenured lecturer position at the University of Jena, where his research focused on transfer and active learning for computer vision, aiming to improve generalization when training data is scarce. Earlier international research experience was gained during a postdoctoral stay at the University of California, Berkeley in 2012–2013, where he worked on domain adaptation and open-set recognition. Between 2018 and 2020, Erik Rodner was part of ZEISS Corporate Research as Machine Learning Lead. In this role, he led work on learning-based methods for optical inspection as well as medical and microscopy image analysis, bridging academic research and industrial applications. Erik Rodner has authored more than 100 peer-reviewed publications in machine learning and computer vision, published in leading international conferences and journals such as CVPR, ECCV, ICCV, TPAMI, and IJCV.

Prof. Dr. Erik Rodner
Professor @ University of Applied Sciences Berlin, Germany

Title: From Intent to Image: A New Paradigm for Color and Texture

Abstract: What if color grading could begin with language instead of reference images? This keynote presents a new approach to visual creation, where intent-based AI transforms simple descriptions into precise, production-ready looks. Alongside this, a purpose-built color space for film emulation is introduced, uniting logarithmic exposure with perceptually accurate color to deliver richer, more natural results. The talk also explores how superpixel-driven texture analysis can be reimagined as texture-aware LUTs, enabling more intelligent and context-sensitive image stylization. Together, these ideas point toward a shift in how creative intent is expressed, translated, and realized in digital imaging.

Bio of the speaker: Dado Valentic is a color scientist and R&D technologist with over 20 years of experience supporting studio production pipelines. His work spans AI-driven color automation, perceptual color science, and imaging systems for feature film and episodic workflows. He collaborates with studios and engineering teams on scalable color pipelines and contributes to industry standards, including ACES and OpenTimelineIO.

Dado Valentic
Colour Scientist and R&D technologist @ Colourlab Ai, US

Title: Connecting the dots: Vision science and the art of magic

Abstract:

Considering that illusions play a pivotal role both in vision science and in the art of magic, it is natural to conceive of these two fields as intimately related, with a huge potential for mutual knowledge transfer. I illustrate how research on amodal completion is of interest to magicians, because it neatly accounts for the surprising deceptiveness of many magic tricks. Using research on the illusion of absence as an example, I also illustrate how we as vision scientists can advance vision science by studying what magicians do and why it works. The illusion of absence, which my collaborators and I stumbled upon while analysing magic tricks is not only of interest from the point of view of basic vision science. Recent work suggests that is also may be a contributing factor in traffic accidents involving blind spots, such as those created by the pillars next to the windscreen.

Bio of the speaker:

Vebjørn Ekroll received his PhD from the University of Kiel and is a professor of psychology at the University of Bergen. His research on various topics such as color perception, motion perception and amodal completion is motivated by a general interest in the basic principles underlying perceptual processes. In recent years, he has focused on identifying hitherto unknown principles of perception and cognition by studying what magicians do and why it works. He is currently interested in how scientific insights from the study of magic can be generalized and translated into applied domains, such as education and traffic safety.

Prof. Dr. Vebjørn Ekroll
Professor @ University of Bergen, Norway

Title: Dynamics in Material Appearance

Abstract:

Humans can visually assess the material qualities of objects, like stickiness, softness, or glossiness, with great ease and speed. Research on the visual perception of materials and objects focuses on understanding the link between visual cues and appearance. Most of this work has involved stationary observers looking at static scenes. Yet, in everyday life we almost always interact with the objects whose properties we wish to assess. These interactions generate dynamic visual input, linking a specific pattern of image motion to the experience of a particular material quality. Several studies, including those from our own laboratory, have demonstrated that image motion can serve as an important cue for signaling material properties, such as softness, shininess, or iridescence. In this talk I will explore the possible connections among the ways we interact with an object, the resulting dynamic visual information, and the material qualities we perceive visually.

Bio of the speaker:

Katja Dörschner received her PhD in 2006 in Experimental Psychology from New York University, USA. After a post‑doc period at the University of Minnesota, she started a position as Assistant Professor at Bilkent University, Türkiye, in 2008. In 2014 she received a Humboldt‑Foundation Sophie Kovalevskaja Award and funding to move to Giessen and establish a research group at the Justus‑Liebig‑University (JLU). At JLU she was appointed Professor in 2019. She is currently coordinating a European Marie Skłodowska‑Curie Doctoral Network focusing on the Perception of materials, objects and spaces through active EXPLORAtion (EXPLORA). In her research, Dr. Doerschner combines psychophysics with tracking methods, computational approaches, virtual reality, and fMRI to uncover and understand the mechanisms that enable humans to perceive intrinsic object qualities, such as material and shape.

Prof. Dr. Katja Dörschner-Boyaci
Professor @ Justus Liebig University Giessen, Germany