AMLD - Applied Machine Learning Days

Had a wonderful experience at the AMLD - Applied Machine Learning Days conference in Lausanne! I had the pleasure to present together with my colleague Adrian Schumacher some of the work we are doing regarding connectivity and AI. Thank you to the organizers Joana Soares Machado and Daniel Dobos for the invitation, and to all the participants that were present! #AMLDEPFL2025 #AI #Innovation #Swisscom

Once again AMLD - Applied Machine Learning Days delivered! A few take-aways from me: 𝟭. 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀 (𝗙𝗠𝘀) 𝗻𝗼𝘄 𝗶𝗻 𝗰𝗼𝗺𝗺𝗼𝗻 𝘃𝗼𝗰𝗮𝗯𝘂𝗹𝗮𝗿𝘆, 𝗻𝗼𝘁 𝗼𝗻𝗹𝘆 𝗟𝗟𝗠𝘀:    • also for geospatial data to apply to weather forecasting    • for electrical grid optimization 𝟮. 𝗦𝗼𝗺𝗲 𝗯𝗶𝗴 𝘁𝗼𝗽𝗶𝗰𝘀:    • by size of room: (gen)AI for finance, (gen)AI for clinical care, AI for energy system    • still a lot of RAG presentations with common pitfalls delivering it 𝟯. 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻𝘀 𝗯𝘆 𝗬𝗰𝗼𝗺𝗯𝗶𝗻𝗮𝘁𝗼𝗿 𝗽𝗮𝗿𝘁𝗻𝗲𝗿:    • high quality test data is a big asset, vertical AI to be bigger than SaaS, voice AI to be mainstream 𝟰. 𝗖𝗮𝘂𝘁𝗶𝗼𝗻 𝗮𝗿𝗼𝘂𝗻𝗱 𝗮𝗴𝗲𝗻𝘁𝘀 𝗮𝗻𝗱 𝗯𝗶𝗴 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸𝘀    • no concrete agentic use case presented - just research stage    • big FSI holding back on going agentic and recommend limit GenAI frameworks when starting (I agree) 𝟱. 𝗢𝗻 𝗴𝗲𝗻𝗔𝗜 𝗮𝗱𝗼𝗽𝘁𝗶𝗼𝗻:    • get champions on the ground not only higher-up to support and promote use-cases    • some features like assisting content generation take off even with not much publicity, users naturally try them 𝟲. 𝗚𝗲𝗻𝗔𝗜 𝗰𝗼𝗺𝗶𝗻𝗴 𝗮𝗹𝘀𝗼 𝗶𝗻 𝗵𝗶𝗴𝗵𝗹𝘆 𝗿𝗲𝗴𝘂𝗹𝗮𝘁𝗲𝗱 𝗲𝗻𝘃𝗶𝗿𝗼𝗻𝗺𝗲𝗻𝘁𝘀:    • hospitals have some use-cases in prod (summarize conv between physcian and patient, public chatbot) 𝟳. 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻 𝘁𝗼𝗽𝗶𝗰 𝗶𝘀 𝗴𝗿𝗼𝘄𝗶𝗻𝗴:    • 2y comparing behaviors of models and setups in CHUV together with physicians    • extensive red-teaming in place at SwissRe    • evolution of economic and social values of different GPT versions 𝟴. 𝗦𝘁𝗶𝗹𝗹 𝗺𝗮𝗻𝘆 𝗱𝗶𝘀𝗰𝘂𝘀𝘀𝗶𝗼𝗻𝘀 𝗼𝗻 𝗻𝗼𝗻-𝗟𝗟𝗠 𝘁𝗼𝗽𝗶𝗰𝘀 (𝘁𝗵𝗮𝗻𝗸𝗳𝘂𝗹𝗹𝘆!):    • predictive maintenance, weather forecast, train traffic optimisation, energy transition decision support, …

Last week, I had the pleasure of attending the AMLD - Applied Machine Learning Days 2025 at EPFL. After organizing my notes, here are some of the most interesting insights, ideas and messages that stuck with me: ◆ The shifting perception of AI: Marcel Salathé highlighted how society has moved from "it just doesn’t work" to "it will surpass humans in a few years." A remarkable mentality shift! ◆ Investing in innovation: Isabelle Moret shared how Vaud is strategically supporting the startup ecosystem, fostering new opportunities for entrepreneurs. ◆ AI startup explosion: Nicolas Dessaigne discussed how ChatGPT ignited an AI startup boom, enabling companies in the AI space to scale from 0 to 10M$ ARR in just a year, which is 10x faster than before the AI revolution. ◆ XLSTMs vs. Transformers: Sepp Hochreiter explained how xLSTMs can outperform transformer-based models both in computational efficiency and energy consumption, without penalizing the performance. ◆ Driving GenAI adoption: Philip Yazdani emphasized that “lateral ambassadors” (peers who organically share knowledge) are the most effective catalysts for corporate adoption of GenAI tools. ◆ New AI security risks: Andrew Paverd shed light on the new vulnerabilities that GenAI models bring to the table: inference manipulation, model manipulation, and inferential information disclosure. ◆ Importance of simple ideas: Ivan Dokmanić showed how treating seismic wave phase as a probability distribution set new state-of-the-art records in phase detection using ML models. ◆ AI for scientific breakthroughs: Christian Lessig reflected on how, in 2020, AI was dismissed for weather forecasting, yet today, WeatherGenerator (https://lnkd.in/ewZiaHw3) not only outperforms traditional forecasting models in accuracy, but also in speed and compute efficiency. ◆ AI for good: Julia Gottfriedsen demonstrated how OroraTech leverages ML to quickly and reliably detect wildfires from space. ◆ AI in software security & automation: Mathias Payer gave a brilliant talk on the three pillars of software security: testing, mitigation, and compartmentalization. This conference was an incredible reminder of how AI is evolving at an unprecedented pace, transforming industries, science, and our everyday lives. And I went back home with some homework: ◆ Include xLSTMs in the set of models to test next time I evaluate different modeling approaches for time series... Sepp Hochreiter did a very good job explaining the advantages of this new architecture (besides being an amazing speaker!) ◆ Take a deeper dive into WeatherGenerator... yes, I am a meteorology geek! (look at this https://lnkd.in/eYT_z-js for instance) Did you attend AMLD25? If yes, is there anything you would add to my list of insights? #AMLDEPFL2025 #MachineLearning #AI #TechConference

Honored to have spoken at AMLD - Applied Machine Learning Days event in front of some distinguished guests & impressive AI talents from the EPFL - big thank you to Joana Soares Machado & Daniel Dobos for the invitation to speak in the AI for a Connected World track 🎤✨ One insight that really hit home: Technology is just a means to an outcome. Jérôme Koller from die Mobiliar shared how some BUs at his organization no longer request for specific tech as strategic projects, but instead define the outcomes they want to achieve. As AI adoption increases through the journey, it’s interesting to observe that the focus moves from tools to tangible outcomes & impact. It’s a great reminder: AI’s power lies not in what it is, but in what it enables to achieve. No better time to take the next step in the #AI journey, as with the #SwissAIPlatform there’s a secure, Swiss-hosted infrastructure, tooling & frontier AI models giving organizations piece of mind from data leakage concerns. 🚀 Reach out to me if this is what your organization has been waiting for. Swisscom Business #GenAI #LLM

A week later, we're still reflecting on all the great discussions, hands-on learning, and connections made at AMLD - Applied Machine Learning Days . One of the key topics that stood out this year was that #GenAI is a key enabler of embedded insurance, driving one of the biggest growth trends in the industry. At our workshop with AXA, we explored how LLMs can support claim handling, diving into real-world applications of AI. Some key takeaways from the session and discussions throughout the event: • LLMs are already changing insurance, streamlining claim validation, and contract analysis. • GraphRAG and QLoRA are helping make retrieval and fine-tuning more efficient. • AI's real impact comes from solving industry challenges—the tech is just the enabler. A big thank you to everyone who joined us, shared insights, and contributed to the discussions. #AMLD2025

Fantastic to have The Al Alliance team at AMLD EPFL 2024. Thank you to The AI Alliance for your support as a partner.

🚗 Worth driving to Lausanne for AMLD - Applied Machine Learning Days 2025! 🚗 Attending in-person events like this reminds me how valuable real-life interactions are, four days filled with insightful talks, engaging discussions, and inspiring exchanges. Huge thanks to the organizers and speakers for putting together such a fantastic event! The conference covered a wide range of topics, focusing on the applications of machine learning and AI across different domains. Honestly, it’s hard to wrap up all these sessions in a single post! From Prof. Sepp Hochreiter’s exciting talk on xLSTM to sessions on Voice AI, AI in Healthcare, and TinyML, the depth and diversity of discussions were truly impressive. Friday’s workshops and hands-on sessions made the experience even more immersive. If you’re looking for a top-tier AI & ML conference, I highly recommend AMLD - Applied Machine Learning Days especially if you're not too far from Lausanne! Coming from Vienna, I can say it was absolutely worth the trip. Don’t forget to visit Holy Cow while you’re at EPFL enjoy the beer and burgers, and experience the cool vibes 🍔🍻 https://lnkd.in/gfcM6aCs

🚀 What an inspiring week at AMLD - Applied Machine Learning Days 2025! It was an honour to host the Swisscom track "AI for a Connected World" with Daniel Dobos! A big thank you to all the speakers for showcasing innovative #AI applications across industries, including IoT, telco, transportation, and disaster management. Your contributions emphasised the critical role of #Connectivity infrastructure, with insightful discussions on shaping its future - while ensuring that AI remains accessible, ethical, and trustworthy for everyone! 🌍 👏 Rosie McDonald, Ines Knaepper, David Froelicher, Marco Rossetti, PhD, Maran Pereirasamy, MBA, Francesco Calabrese, PhD, Ian Sutcliffe, Stavros Karagiannopoulos, Adrian Schumacher, Gerald Sula, Martin Tefra, and Paul Keller It was also a pleasure to present at the AI & Cybersecurity session - thank you David Froelicher for the invitation! Exciting to see the latest advancements in privacy-preserving AI, security, and AI governance. 💡 Highlights: - Big potential for collaborative open-source innovation and for exploring new techniques to improve model efficiency (e.g., DeepSeek) - levelling the playing field for startups and providing a path towards more sustainable AI. - Agentic RAG: combining LLMs with search and information retrieval techniques, leveraging cutting-edge advances in LLM frameworks. - Interdisciplinary research: taking inspiration from the brain to build more efficient AI systems by Mackenzie Mathis, PhD, and teaching LLMs the "language" of biology by David van Dijk. Thank you to the AMLD organisers for their support! We are already looking forward to AMLD 2026! ✨ #AMLDEPFL2025 #AI #Innovation #Swisscom #PerformingTogether