KEMAI Summer School 2026

Wednesday, July 15, 2026: 9:10 – 10:10, Schloss Reisensburg, Roter Saal

Abstract:
Large language models (LLMs) are increasingly used to automate structured annotations of clinical data grounded in biomedical ontologies. However, such models perform stochastically, are sensitive to context and prompt formulation, and may evolve without accessible version archives. As a result, they are rapidly introducing a reproducibility crisis. This presentation will examine how this problem can be at least partially mitigated, with concrete examples taken from biochemistry, clinical notes, and evidence synthesis in mental health, concluding with highlighting some persistently open research challenges in this area.

Bio:
Janna Hastings was born in Cape Town, South Africa, where she completed her undergraduate studies in Mathematics and Computer Science. Thereafter, she moved to Cambridge, UK, to join the Cheminformatics and Metabolism group at the European Bioinformatics Institute (2006–2015) where she led the development of the ChEBI molecular ontology and metabolism knowledgebase. She completed part-time Master’s degrees in Computer Science (University of South Africa, 2011) and Philosophy (Open University, 2012) before obtaining her PhD in Computational Biology from the University of Cambridge (2015–2019), where she studied the role of metabolism in healthy aging using multi-omics data and a time-series modelling approach. She completed postdoctoral studies at the Otto-von-Guericke University Magdeburg (2019–2022), at the EPFL (2020–2022), and with the Human Behaviour-Change Project at University College London (2017–2022). Between 2022 and 2025 she was an Assistant Professor in the Medical Faculty of the University of Zurich, and since January 2026 she has taken up the position of Senior Research Scientist at the Idiap Research Institute, where she leads the Human-Centered Health AI group focused on advancing AI methods for health-related applications, particularly hybrid approaches that combine knowledge with large-scale data.

Wednesday, July 15, 2026: 11:20 – 12:20, Schloss Reisensburg, Roter Saal

Abstract:
Foundation models have become an important ingredient in medical image analysis, especially in computational pathology. They hold the promise of reducing the need for annotated training data, increasing robustness, and broadening applicability. Still, their real-world value depends on robust adaptation across scanners, clinical sites, and application domains. This talk will examine recent research on domain shifts, benchmarks, and data-efficient model adaptation in medical imaging, with a focus on histopathology. It will explore typical adaptation strategies such as linear probing, end-to-end training, self-supervised approaches, and parameter-efficient methods, showing that foundation models are powerful starting points but are not automatically superior without careful adaptation. Particular focus will be placed on their potential in limited-label settings, while highlighting open challenges in generalization, validation, and domain-aware evaluation.

Bio:
Katharina Breininger is Professor of Pattern Recognition at the Center for Artificial Intelligence and Data Science (CAIDAS) and the Institute of Computer Science at Julius-Maximilians-Universität Würzburg. Her research focuses on robust and generalizable machine learning for interdisciplinary applications, with a particular emphasis on medicine and medical imaging. Her group investigates representation learning, generalization under domain shifts, annotation and labeling strategies, human factors, and human-AI collaboration, while also developing open-source tools and datasets for semi-automatic and collaborative annotation. Application areas include intraoperative and multimodal imaging, interventional imaging, microscopy, and digital pathology. She studied computer science in Marburg and Erlangen and completed her PhD in 2020 at FAU Erlangen-Nürnberg on image fusion for minimally invasive interventions in collaboration with Siemens Healthineers. Before joining Würzburg in 2024, she was W1 Professor of Artificial Intelligence in Medical Imaging at FAU, where she led the AI in Medical Imaging lab.

Wednesday, July 15, 2026: 14:35 – 15:35, Schloss Reisensburg, Roter Saal

Abstract:
My talk traces the intellectual arc from the birth of data driven medical science to modern foundation models in healthcare, using John Snow as a conceptual starting point and extending to our development of Nightingale AI. It examines how the core logic of data-driven inference — linking observations to actionable interventions — has evolved from statistics and causal reasoning to large-scale, multimodal machine learning. We will discuss how contemporary ML systems move beyond static prediction toward continuous patient modelling, integrating heterogeneous data streams (EHR, imaging, biosignals, behaviour) into unified latent representations. The talk will highlight the role of foundation models, reinforcement learning, and related approaches in enabling this transition, as well as the constraints imposed by clinical validation, regulatory frameworks, and real-world deployment. The objective is to provide a principled view of how modern AI systems can be understood as a natural extension of data driven reasoning — while also identifying where fundamentally new methodological and conceptual challenges emerge.

Bio:
Aldo Faisal is a Professor of AI & Neuroscience jointly at the Dept. of Bioengineering and the Dept. of Computing at Imperial College London, where he leads the Brain & Behaviour Lab. Aldo is also Director of the Behaviour Analytics Lab at the Data Science Institute. He is also Associate Investigator at the MRC London Institute of Medical Sciences and is affiliated faculty at the Gatsby Computational Neuroscience Unit (University College London). Aldo serves as an Associate editor for Nature Scientific Data and PLOS Computational Biology and has acted as conference chair, area chair, and program chair in key conferences in the field (e.g. Neurotechnix, KDD, NIPS, IEEE BSN); in 2016 he was elected into the Global Futures Council of the World Economic Forum. Aldo received a number of awards and distinctions, including being a scholar of the German National Merit Foundation (Studienstiftung des Deutschen Volkes), a Fellow of the Böhringer-Ingelheim Foundation for Basic Biomedical Research, elected as a Junior Research Fellow at the University of Cambridge (Wolfson College), and research prizes such as the Toyota Mobility Foundation award in 2018 ($50,000). Aldo’s lab has featured regularly across global media (BBC, CNN, WIRED, TED, TEDx, New Scientist, Guardian, Times of India, etc.) and Scientific American voted his research on gaze-based control as 1st of 10 most transformative ideas for 2016.

Thursday, July 16, 2026: 9:00 – 10:00, Schloss Reisensburg, Roter Saal

Abstract:
Artificial intelligence is increasingly transforming healthcare, ranging from clinical decision-support systems and AI-assisted diagnostics to personalised medicine and the secondary use of health data for research and innovation. At the same time, the growing reliance on data-driven technologies raises fundamental legal and ethical questions concerning data governance, accountability, transparency, patient autonomy and the protection of fundamental rights. The presentation explores the evolving European regulatory framework for AI-driven healthcare, including the interplay between data protection law, health law and AI regulation under the EU AI Act. Particular attention will be paid to the challenges of balancing innovation incentives and technological progress with the need to ensure patient safety, trustworthy AI and effective protection of individual rights, including ongoing debates on the assessment, governance, transparency and regulation of risks associated with AI-enabled healthcare. Against the background of the ongoing digital transformation of healthcare, the presentation will discuss how legal and regulatory frameworks can contribute to responsible innovation and sustainable governance of AI-enabled health technologies in Europe.

Bio:
Prof. Dr. Claudia Seitz is Professor of Public Law, European Law, International Law and Life Sciences Law and Head of the Institute of European and International Law (IEIL) at the Faculty of Law of the Private University in the Principality of Liechtenstein (UFL). She also holds academic appointments as Visiting Professor at Ghent University (Belgium) and as Senior Lecturer at the Universities of Basel (Switzerland) and Bonn (Germany). Her research focuses on European Union and international law, health law, digital health regulation and the legal governance of emerging technologies, in particular artificial intelligence. She holds, i.a., an Executive Master in Law and AI from the Brussels School of Competition Law. Prof. Seitz has been affiliated with several international research initiatives in the fields of biotechnology and life sciences law. She is co-editor of the European Health and Pharmaceutical Law Review (EHPL) and the Europäische Zeitschrift für Wirtschaftsrecht (EuZW). In addition, she serves as advisor to international organisations, including the World Health Organization (WHO), the World Intellectual Property Organization (WIPO) and the EU AI Office of the European Commission.

Thursday, July 16, 2026: 10:15 – 10:45, Schloss Reisensburg, Roter Saal

Abstract:
Deep neural networks (DNNs) achieve unbeaten performance in automated processing of non-symbolic sensory inputs like images. This includes safety-critical application areas such as automated driving or diagnosis assistant systems in healthcare. Such domains come along with a rich set of prior, human-defined symbolic (expert) knowledge, like known object relations (e.g., a head typically indicates a pedestrian) or medical expert knowledge (e.g., a skin cancer region is usually colored different from its surroundings). System safety certification now demands strict compliance to these prior rules, making it a compelling goal to use available knowledge directly during DNN training, peek inside the DNN to check whether the knowledge was learned correctly, and potentially correct errors. These aims face the opacity and distributed representations of DNNs as a major difficulty: one cannot directly read off which knowledge was learned about human-interpretable concepts and their relations, hindering integration, extraction, (formal) verification, and correction of DNNs against symbolic knowledge. The field of concept-based explainable AI (C-XAI) poses a promising pathway to bridge this gap, associating human-readable specifications with latent encodings in DNNs. In this talk I introduce the basics and pitfalls of concept-based explainability, and give a peek into my research on novel C-XAI explanation techniques as well as their application to verification and correction of computer vision DNNs.

Bio:
Gesina Schwalbe started her tenure track juniorprofessorship for Explainable Artificial Intelligence in April 2026 at the Institute of Artificial Intelligence, Ulm University. Here she continues her lead of the junior research group Correctable Hybrid Artificial Intelligence, for which she acquired funding starting in Sep 2024 during her postdoctoral time at University of Lübeck (2024–2026). Her group investigates explainable AI techniques for the verification, correction, and integration of symbolic knowledge into deep neural networks, with applications in automated driving, chemistry, and medical AI. This directly builds upon her research on explainability for safe and trustworthy AI, which she conducted as project lead (2021–2023) and doctoral researcher (2018–2022, supervised by Prof. Ute Schmid, Otto-Friedrich-Universität Bamberg) at the Continental AG automotive supplier.