Marina Hader

From hype to impact: How AI is reshaping healthcare The momentum from Applied Machine Learning Days 2026 continues to inspire how we think about AI at Roche. At AMLD, we explored one central idea: 👉 The future of AI in healthcare is not about bigger models, but about biology-aware, patient-centric intelligence. What stood out: 🔬 Reimagining drug discovery From multimodal antibody design to multi-agent systems accelerating target evaluation: AI is becoming a true research partner, deeply grounded in biology. ⚙️ Scaling AI responsibly With over 1,200 active AI agents on our enterprise platform, we are democratizing AI across the value chain: from research to development. 🧬 Transforming clinical development Digital Twins, biology-informed models, and AI-powered endpoints are helping us de-risk trials, increase precision, and move faster for patients. 📊 Turning data into decisions From wearable-derived disability scores in Multiple Sclerosis to AI-based tumor measurements: we are bringing objective, clinical-grade insights into practice. AI in healthcare is maturing 👉 from promise to real-world impact. A big thank you to everyone who contributed to these inspiring discussions. Luca Finelli, Prasoon Paliwal, Asif Jan, Elif Ozkirimli, Tiba Razmi, Fethallah Benmansour, Frédéric Dreyer, Markus D. Herrmann, Vittorio Paolo Illiano, Thomas Zaugg, Raul Rodriguez-Esteban, Stanislaw Adaszewski, Stefan Keel 📸 Photo: Samuel Devantery #AMLD2026 #Roche #Genentech #HealthcareAI #DrugDiscovery #ClinicalTrials #DigitalHealth

158

2 Kommentare

AI-ready actuaries treat #AI as informative—not authoritative. Judgment, humility, governance and peer review remain essential. Find out more from the expert panel report: https://bit.ly/3MIaIga #actuaries #actuaryresearch Author: Ronald Poon Affat

9

Henning Von Kielpinski, Head of IT Underwriting Solutions at Hannover Re, captivated the audience with his session, “From Steam Engines to LLMs: Why AI Came to Stay”. He navigated the complexities of AI in regulated environments, urging organizations to "tread lightly" while daring to "fail fast". By sharing real-world use cases that simplify the lives of underwriters, Kielpinski proved that AI is no longer just a tool, but a fundamental shift in operational reality. His insights provided a roadmap for balancing high-stakes compliance with cutting-edge innovation. #ICEM2026 #HannoverRe #Insurtech #GenerativeAI #LLM #FutureOfWork #AIStrategy #RegulatedInnovation

8

🎉Thrilled to see the KODAQS tools published in Die Publizistik! https://lnkd.in/dSU2x9Ny Honoured to be part of the toolbox team! 🙌. Special thanks to Jessica Daikeler, Katrin Weller, and Yannik Peters for the constant support, mentorship, and for bringing me into this incredible collaboration. My contribution focused on digital behavioural data, where I worked on the TextPrep tool together with Yannik Peters. It explores how different text preprocessing steps impact data quality using simulated social media data from the 2024 Summer Olympics. You can access the TextPrep work here: DOI [https://lnkd.in/dYzwX2x8] or via Google Scholar [https://lnkd.in/dnVYv8mz] Excited for what's ahead as we enter the final project phase and more tools ahead! 🚀 #DataQuality #DigitalBehavioralData #TextPreprocessing #KODAQS

16

New in CODATA Data Science Journal: Sonja N. Kralj, Matthias Landwehr Title: Research Data Management Practices Across the Research Data Lifecycle and Their Potential for Collaboration in an International Higher Education Alliance https://lnkd.in/eRMCDcjy #codata #FAIRdata #OpenScience #datascience #opendata #datasciencej

10

I was happy to share my perspective with Frankfurter Allgemeine Zeitung for their recent article on AI manipulation. It highlights a critical challenge of prompt injections https://lnkd.in/epcHxZzf   While the risks are real and scary, and 100% security is unattainable in any system, they can be reduced by following best practice.   Prompt injections are just one part of how AI is reshaping (IT) security. AI can mean more effective, automated, accelerated attacks - but it can also help protect systems better. At the University of Kaiserslautern-Landau, my team and I even explored the educational potential of these techniques, using hidden white text in an exam to see if students in our "Engineering with Generative AI" course were using AI for their answers. (Islam Mesabah)   The key is to foster a nuanced understanding, recognizing the vulnerabilities while harnessing AI's transformative power responsibly.   If you are concerned about AI's impact on security or harnessing AI, get in touch. We can help.   #AI #Cybersecurity #PromptInjection #Innovation #DFKI #RPTU

60

1 Kommentar

Thrilled to share how we helped LVM Versicherung, a leader in the German insurance industry, modernize its digital presence! 🤕 𝐓𝐡𝐞 𝐂𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: LVM needed to provide a better online experience to support its customers and network of local agents, but its old on-premise CMS was too inflexible and inefficient. 🎯 𝐓𝐡𝐞 𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧: By implementing Magnolia's composable DX Cloud, LVM now has the power and flexibility to deliver personalized experiences, accelerate content updates, and easily integrate new tools. This success story is a testament to the power of moving from a restrictive legacy system to a flexible, cloud-first solution. Their teams are now empowered to connect with customers more effectively than ever before. Want to learn more about how we drive digital transformation? 👉 Explore the full case study: https://brnw.ch/21wWpLl #SuccessStory #DigitalMarketing #Insurance #Cloud #ClientSuccess #MagnoliaDXP

25

What if your most expensive actuarial talent spends too much time… on low‑value work? In many German insurers, actuaries sit at the heart of: * Solvency II reporting * pricing and reserving * risk and capital management Yet in practice, a large share of their time still goes into: * data reconciliation across legacy systems * manual validation steps * reporting cycles that can take weeks 👉 In a BaFin / DORA environment, this is more than inefficiency. It creates delays, audit complexity and operational risk. This is where the idea of an “Agentic Actuary” becomes interesting. The Kyndryl use case shows how AI agents embedded across actuarial workflows can: * automate end‑to‑end reporting and validation processes * generate regulatory outputs and perform continuous compliance checks * enable near real-time insights for pricing, reserving and risk steering At the same time: human actuaries shift from data processing to judgement, scenario steering and risk strategy Why this matters for German insurers: * BaFin expectations require traceability, auditability and control * DORA demands resilient, well-governed digital processes * Skills shortages in actuarial roles are increasing pressure 👉 AI is not replacing actuarial expertise — it is becoming the lever to scale it safely and compliantly. My takeaway: The real value of AI in insurance is not automation alone. It’s the ability to: => combine speed, control and regulatory alignment => while unlocking scarce expertise for higher-value decisions 👉 Explore the use case: http://ms.spr.ly/6045vXNX1 How are you currently approaching the balance between efficiency, auditability and actuarial capacity? #Insurance #Germany #Actuarial #BaFin #DORA #AINative #AgenticAI #RiskManagement #Kyndryl #TheHeartOfProgress

5

Harald Trautsch once said: “The point is not to provide more information. The point is to provide the right information.” In the digital insurance world, bombarding customers with constant updates, alerts, and details can backfire. Why? Because information overload dilutes value — and customers tune out. At Dolphin Technologies, we believe the future belongs to insurers who know how to deliver the right information, at the right time, in the right context. ✅ Prioritize relevance over volume ✅ Align messaging with user needs and moments ✅ Create smarter, not louder, communication And don’t just talk about timing — we build it. Our MOVE SDK enables real-time, context-aware communication tailored to how people actually drive. Check it out: https://lnkd.in/d77d7Fbk 🎧 Dive into this discussion on Episode 6 of the Insurance Telematics Podcast: Watch or listen here: YouTube: https://lnkd.in/dnDzS96t Spotify: https://lnkd.in/dERqiQRn Apple Podcasts: https://lnkd.in/dKrD-5kS

27

1 Kommentar

Cool interview (in German 🙃) with Eric Upschulte about our contour-based object detection models for quantifying brain architecture from extremely detailed microscopy images as provided by our BigBrain Project and through EBRAINS. Besides the powerful AI models, Eric developed a well designed software library and interfaces to common tools like napari - check out at https://celldetection.org, it rocks! 🚀 (Soon, model predictions will also be accessible in realtime through our siibra toolsuite, e.g. at https://lnkd.in/er2BNfD2 when viewing zoomed-in brain images…)

24

The Actuarial Association of Europe Education Committee has recently published important 𝗻𝗲𝘄 𝗴𝘂𝗶𝗱𝗲𝗹𝗶𝗻𝗲𝘀 𝗼𝗻 𝗔𝗜 𝘀𝗸𝗶𝗹𝗹𝘀 tailored specifically for actuaries. ⬇️ In this must-watch video, Rafael Moreno Ruiz and Olivier Lopez take you through these guidelines, explaining which AI competencies matter most in actuarial work and how associations can design effective CPD programs around them. The session then dives deeper into Generative AI – exploring its implications for actuarial education and how this cutting-edge technology can be integrated into training to keep actuaries ahead of the curve. 🔗 Don’t miss out on how AI is shaping the future of the actuarial profession and how to prepare for it – watch now: https://lnkd.in/ex4-NAMT #AI #AIforActuaries #CPD #GenAI #GenerativeAI

9

#TimeUseSurveys are conducted to understand how people allocate their time, inform evidence-based policymaking, and shed light on social, economic, and gender-related patterns. In contrast to previous rounds, the 2022 German Time Use Survey introduced a digital diary tool (#Zeitverwendung app, ZVE app), allowing respondents to record their daily activities using either a smartphone app or a web-based application. Statistisches Bundesamt (Destatis) has now published some first methodological findings from the #ZVE2022-study designed around an online-first approach. To explore the main insights in detail, I recommend reading the full article published in WISTA – Wirtschaft und Statistik(Issue 5/2025) or, at the very least, reviewing the key takeaways summarized in the hbits post below. Source: https://lnkd.in/eEAZWPCC #IATUR #timediary

11

📣 I am very happy to announce that Maria-Esther Vidal, Cinzia Cappiello, annette ten teije, Tarek R. Besold and I will organize a seminar at Schloss Dagstuhl – Leibniz-Zentrum für Informatik (LZI) on "Research Data Management in Times of Agentic AI" from May 30 – Jun 04, 2027. Generative AI, especially LLMs, is evolving into agentic AI systems that can plan, reason, and act across the research lifecycle. This brings major opportunities, but also raises challenges for research data management (RDM), particularly around reproducibility, transparency, and accountability. This seminar will explore how RDM infrastructures, practices, and governance must evolve to support the responsible use of agentic AI in science and industry, with a focus on semantic technologies and neuro-symbolic AI for machine-actionable, explainable, and trustworthy research. 🎯 Key questions discussed in the seminar include: * Which FAIR RDM tasks can be responsibly delegated to agentic AI and under what conditions? * How can we ensure agentic outputs are trustworthy, explainable, and accountable to researchers and institutions? 📚 We will organize discussions along multiple topics comprising: * FAIR and Structured Knowledge for Agentic AI Towards Machine-Actionable Research * Trustworthiness in Agentic AI for Research * User-Centric Agentic AI for Supporting Research * Applications of Agentic AI-driven Research 🔭 The seminar aims to bring together researchers from AI, databases, Semantic Web, research infrastructures, libraries, and industry to develop a shared research agenda and a roadmap for trustworthy, FAIR-aligned, explainable, and adaptive RDM in the age of agentic and neuro-symbolic AI. 🔗 Further seminar details can be found here: https://lnkd.in/eAdgYJz3

82

2 Kommentare

A highly recommended read for anyone who wants to move beyond the buzzwords around Artificial Intelligence. In this blog post (text in German), Axel Polleres — Key Researcher at BilAI and Member of the Board of Directors — provides a clear and engaging explanation of what AI is, what it (still) cannot do, and how it actually works. 👇

8

#SeedFundSpotlight      Researchers from the University of Tübingen Department of Computer Science and the UNC School of Data Science and Society are working together to solve an ongoing challenge for children with epilepsy.    An epileptic seizure can be fatal if it occurs during sleep. Through funding from the UNC-Tübingen Seed Fund for Research Collaborations, Harlin Lee (UNC) and Mete Akgün (Tübingen) have been advancing their project "Privacy-Preserving Machine Learning Framework for Early Detection of Pediatric Epilepsy" to combine data science, artificial intelligence and pediatric medicine to develop secure and efficient models for real-time prediction of pediatric epilepsy during sleep.    The project combines Akgün’s expertise in privacy-preserving machine learning with Lee’s research on large-scale pediatric sleep data, aligning The University of North Carolina's strengths in federated learning with Tübingen’s leadership in biomedical data privacy. Using the NCH Sleep dataset, which includes over 3,900 pediatric sleep studies, Akgün and Lee are training models that can analyze data collaboratively without revealing patient information — simulating real-world clinical scenarios while maintaining both privacy and accuracy.     In 2024, Akgün and Lee held bi-weekly virtual meetings, conducted algorithm development and testing, and continue to exchange visits between Tübingen and Chapel Hill to expand interdisciplinary collaboration. They are planning to submit their project findings to a computational medicine journal, which they hope will serve as a foundation for future grant proposals, graduate training and large-scale research initiatives related to #AI and health data science.     This project highlights the power of international collaboration to advance secure, ethical and impactful AI for health-related issues — demonstrating how global partnerships are driving innovation in data science and precision health.     Are you working on a project that has — or could benefit from — transatlantic collaboration? The fall 2025 seed fund project proposal deadline is Nov. 15.‼️   ➡️ Learn more and apply on our website: https://lnkd.in/e7Fuhm6m

29

1 Kommentar

Short science talk at Hamburg University of Technology (TUHH): ‘Concept Drift and Hierarchical Algorithmic Solutions in Federated Learning.’ Together with Hamburg 3-Minute-Thesis 2025 finalists and fellow research colleagues from TUHH, I will be sharing the stage once again, this time as part of the talk series: “What’s Your Research About?”, organized by IEEE & VDE Student Group at TUHH. In concise 3–5 minute talks, the speakers will shed light on their research topics across disciplines. other Talks include: Charlotte Goblirsch (Institute of Innovation Marketing, TUHH): Exciting, Green… and Still Ignored: The Hidden Struggle of Sustainable Delivery Mohsen Falah (Institute for Metal and Composite Structures, TUHH): First Aid for Our Steel Infrastructure: Unleashing the Power of Nanotech #TUHH #IEEE #3MT #ResearchCommunication #InterdisciplinaryResearch #FederatedLearning #ConceptDrift #ScienceCommunication

44

2 Kommentare

Interview with Christin-Isabell Henn (COO) and Claude Eisenmann (Director of Software and Data Engineering) from Pax Versicherung. Learn why technology, software and data are important for a mid-sized insurance company, and how this aligns with the values and the spirit of the BaselHack (baselhack.ch).

66

𝗔𝗜 𝗶𝗻 𝗖𝘆𝗯𝗲𝗿𝘀𝗲𝗰𝘂𝗿𝗶𝘁𝘆: 𝗛𝘆𝗽𝗲, 𝗛𝗼𝗽𝗲—𝗼𝗿 𝗮 𝗥𝗲𝗮𝗹 𝗚𝗮𝗺𝗲-𝗖𝗵𝗮𝗻𝗴𝗲𝗿? In a presentation at Bundesamt für Rüstung armasuisse, Daniel Buehlmann demonstrated how AI is now being used consistently on both sides – by attackers as well as in defense. As attacks become faster, more sophisticated, and more widespread, attackers operate without rules, thereby gaining a structural advantage. This is precisely why it is clear that AI and modern technologies must also be used strategically on the defensive side. The key is not to deploy as many tools as possible, but rather the right ones – tailored to one’s own organization, thought out holistically, and meaningfully integrated. Because often, even in cybersecurity, less is more. #Cybersecurity #AI #ArtificialIntelligence #SecurityStrategy #Security #baseVISION

20