ETH AI Center

🚨 AI+X Summit: Switzerland’s largest AI ecosystem event returns on 2 October at StageOne Oerlikon, Zürich.🚨 Organized by the ETH AI Center in collaboration with ZHAW Zurich University of Applied Sciences and University of Zurich, 🎟️ Early-bird 25 % off until 31 July with code #SUMMIT25 🔗 Tickets: https://lnkd.in/gdE9tQvC What to Expect 🔹 Two Stages • Main Stage – high-impact keynotes & panels by global thought leaders, researchers, and industry pioneers • Innovator Stage – rapid-fire talks focusing on entrepreneurship and applied AI 🔹 Curated Tracks & Workshops • AI + Finance, AI + Healthcare, AI in Education, AI for Science, SMEs AI Upskilling, AI + Quantum, Machine Perception for Human Understanding, Efficient LLM Fine-tuning and more sessions to be announced soon! 🔹 Demo Zone • Meet 50+ partners showcasing showcasing AI-driven innovations across AI sectors. 🔹 Networking Tool   • Chat, request meetings and plan face-to-face in our on-site Networking Lounge via the AI + X app 🔹Confirmed Partners (and Growing!) ZHAW Zurich University of Applied Sciences · University of Zurich · Meta · The Applied-AI Center @ HSLU · Generali Switzerland · Beekeeper · Zalando · Inditex · D ONE – Data Driven Value Creation · Founderful · Wenger Vieli · Microsoft Innovation Hub Switzerland· National University of Singapore · IBM Research · …and more joining soon! ▶️ Watch the #plusxsummit2024 Recap: https://lnkd.in/gV59m-xp 🔔 Secure your ticket and join 2000+ AI leaders, researchers and innovators under one roof. Early-bird ends 31 July! Alex Ilic Letizia Turri Valeria Bella Helga Rietz Arnout Devos Dr. Ann-Lauriene Schumacher (Haag) Daniel Naeff ✨ Nicole Barth Natalia Marciniak Trinh Ngo Mauricio Moncada Alicia Rieckhoff Andreas Krause Andreas Henrici Markus Christen Thilo Stadelmann Ruben Kranendonk Fanny Lehmann Manuel Kaufmann Imanol Schlag Bingcong Li Kai Lion

From 18–31 August, we’re running Pokémon AI Adventure at ETH Zurich, a playful 1-hour workshop where kids aged 10–16 get to teach a friendly robot how to classify Pokémon, explore the world of Novara, and pick up the difference between rule-based algorithms and data-driven algorithms along the way. No prior experience needed and the workshop is conducted in both German and English. Parents are welcome to watch and join in the fun! 📍 ETH CAB Building (in person only) 📅 18–31 August

𝗦𝗰𝗮𝗹𝗲 𝗬𝗼𝘂𝗿 𝗔𝗜 𝗦𝘁𝗮𝗿𝘁𝘂𝗽 𝗔𝗰𝗿𝗼𝘀𝘀 𝗘𝘂𝗿𝗼𝗽𝗲! Ready to scale your AI startup with the best support from Europe’s top innovation hubs? Join AI Launchpad 2.0, a curated program connecting you to markets, investors, and experts across 10 leading AI ecosystems. ✨ 𝗪𝗵𝗮𝘁’𝘀 𝗶𝗻 𝗶𝘁 𝗳𝗼𝗿 𝘆𝗼𝘂?  • Tailored 1–2 week immersion visits to premier AI hubs  • Access to a pan-European network of AI entrepreneurs and experts  • Strategic connections with investors, corporates & startups 📅 𝗗𝗲𝗮𝗱𝗹𝗶𝗻𝗲: 𝗔𝘂𝗴𝘂𝘀𝘁 𝟯𝟭, 𝟮𝟬𝟮𝟱, 𝗮𝘁 𝟮𝟯:𝟱𝟵 𝗖𝗘𝗧 👉 𝗚𝗲𝘁 𝗻𝗼𝗺𝗶𝗻𝗮𝘁𝗲𝗱 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗙𝗮𝗹𝗹 𝗦𝗲𝗮𝘀𝗼𝗻 𝗻𝗼𝘄! Learn more & apply: https://lnkd.in/dpxn-A6a Questions? entrepreneurship@ai.ethz.ch #AILaunchpad #AIEurope #StartupScaleup #TechInnovation #AIStartups #EuropeanAI #ELIAS #HorizonEurope INiTS | Vienna's High-Tech Incubator, DTU Skylab, Campus Founders, ETH AI Center Talent Kick, KTH Innovation, XPLORE, Hasso Plattner Institute, Computer Vision Center, YES!Delft, Cyber Valley

The 2nd Call for Large Grants is NOW OPEN! 🚀 We’re excited to launch the next phase of the hashtag #SwissAIInitiative, providing funding and compute resources for cutting-edge research in foundation model development and high-impact application areas. Compute: Up to 15M GPU hours available Financial contribution: ~CHF 2M in personnel & research support Project Duration: 1 year Key dates to keep in mind: Letter of Intent: Aug 25, 2025 Full Proposal: Sept 8, 2025 Full guidelines and submission details are available here: https://lnkd.in/e862trt5 For any questions or clarifications, please do not hesitate to contact grants@swiss-ai.org We look forward to receiving your proposals!

It was a pleasure to present our research at the prestigious Academy of Management Conference 2025 in Copenhagen, where Prof. Dr. Gudela Grote (Professor of Work and Organizational Psychology, ETH AI Center, D-MTEC ETH Zurich), Dr. David Grüning (Postdoctoral Researcher Stanford University and Max Planck Institute for Human Development) and I (ETH AI Center, D-MTEC ETH Zurich), presented our pilot study "Exploring AI Developers Social Responsibility: The Role of Working Conditions and Uncertainty Mindset" within the symposium "(Tech) A New Hope: Toward a More Nuanced Understanding of Workplace Technology Use and Employee Well-Being" organized by Ryan Cheng (The University of Sydney) and Isabelle de Roche (King's College London). In this exploratory survey conducted with AI developers recruited via Prolific, we assessed job satisfaction, job autonomy, social responsibility in AI, their responses to an AI design scenario—choosing between bias-preserving and bias-transforming metrics, and how they appraise the multiple sources of uncertainty in their work—using the Uncertainty Mindset scale, distinguishing between viewing uncertainty as enabling versus disabling (developed by Federico Magni (Nanyang Business School) et al,. including my colleague Alina Gerlach, ETH Zürich, and Prof. Grote) Our framework draws from Prof Sandra Wachter 's (Oxford Internet Institute, University of Oxford) classification of bias preserving metrics vs bias transforming metrics under EU non-discrimination law. Our preliminary findings suggest that high job autonomy may act as a double-edge-sword, being associated with high social responsibility and a preference for bias-preserving metrics at the same time. Even highly socially responsible AI developers may use bias-preserving metrics, leading to AI systems that replicate historical bias—potentially resulting in reputational damage and prima facie discrimination lawsuits under EU non-discrimination law. Requiring explicit AI ethics training and a strong ethical culture balanced with job autonomy. Finally, that encouraging developers to view uncertainty as an opportunity for growth and learning—an uncertainty-as-enabling mindset— may enhance job satisfaction and may encourage reflection on the social and ethical uncertainties and considerations in AI development by reducing resistance to unknown challenges. Pre-print coming soon!

⚙️ Yesterday, we had the pleasure of having Dr. Uma Rani from the International Labour Organization give a talk on "The Human Cogs in the AI Machine" in our AI Center Academic Talk Series (AICATS). 🙏 Thanks to our ETH AI Center members Dr. Arlette Danielle R. (Associated Researcher) & Prof. Dr. Gudela Grote (Faculty member) for hosting them! 🔥 More information on our academic talks, including proposing one for Fall 2025: https://lnkd.in/dnmqXXFN

Slides of our #ACL2025 tutorial "Large Language Models for Education” where we discuss current challenges of LLMs, evaluation metrics, and promising directions to train your own models: https://lnkd.in/ebiSUqq6

Post #ICML post, a bit overdue. I wanted to signal our two recent papers to the continual learning community: you may want to add these to your post-ICML reading list! 📚 We studied continual learning from the perspective of learning regimes, and in these two papers we ask two fundamental questions. ❓ First question: is feature learning beneficial in continual learning? ⏱️ Short answer: no, unless the tasks are all very similar.  📖 Long answer: feature learning has been shown to be a key ingredient in the performance gains that neural networks achieve when compared to kernel methods. On a single task, the implicit bias contained in the feature learning dynamics yield better generalization and transfer and faster convergence. However, the same gains are not visible in continual learning, where the training distribution is non-stationary. In this work, with an elegant experiment design and the powerful tools of dynamical mean-field theory we show exactly how much feature learning is adequate in non-stationary training. We find that feature learning is increasingly harmful as the task similarity is decreased, and that laziness instead shields the network from the disruptive changes in task. This tells us that the same implicit biases which are helpful in a single task setting, exacerbate the network's predisposition to catastrophic forgetting during continual learning. What then are better implicit biases for continual learning?  🔗 You'll find 9 pages of discussion and many more insights at https://lnkd.in/eGVqb84E, or simply reaching out to the talented authors Jacopo Graldi, Alessandro Breccia, Thomas Hofmann, Lorenzo Noci. ❓ Second question: are the fixed NTK analyses valid in continual learning? ⏱️ Short answer: not in practice! 📖 Long answer: Under certain conditions the NTK stays constant during training, yielding an effectively linear function behavior whilst keeping a special neural network inductive bias. This linear form is convenient as it's mathematically tractable and was used by several works to analyze network dynamics in continual learning. However, it is not clear whether the NTK would remain static under a distribution shift, as most proofs assume training on a stationary data distribution. In this work we ran a couple of experiments to answer this simple question, and we observe that in all the settings we tested, regardless of the parametrization, the NTK spikes at the new task onset, a phenomenon which we call "reactivation". Curiously, we observe "reactivation" only when the new task introduced new classes, and not when it introduced a domain shift.  🔗 You'll find the paper here: https://lnkd.in/eV8G_n6y This work was done in collaborations with some great students Yuzhi LIU, Zixuan Chen, Zirui Zhang, Yufei Liu.

 📨 ONE DAY LEFT. One application form. One decision that could shape your future. 🕐 Calling all aspiring entrepreneurs  It’s the final day before our application portal closes! 🕐 ✍ Have you sent us your application yet? It doesn’t take long, and it might change the trajectory of your life! 🌀 📨 If you’re a student or researcher at a Swiss university with an entrepreneurial drive and vision to create positive change — this is an opportunity you should not miss! 🎉 At Talent Kick, we bring together the best talents from across Swiss universities to form strong, interdisciplinary co-founder teams. We are here to support you to gain hands-on entrepreneurial experience, mentorship, and personal coaching during your studies to develop your own startup! Apply today, ONLY ONE DAY LEFT! 🌱 Early Bird Application – TK-S26 (Spring 2026 Batch) – Deadline: August 3rd 23:59 (CEST). Talent Kick will help you build a cross-university co-founder team and work with other entrepreneurial leaders. You will have the chance to boost your project with 5’000 CHF, benefit from study synergies, network access, and support until you graduate. 👀 Immersion Day on September 22nd – Deadline: August 3rd 23:59 (CEST). An open one-day deep dive into the Talent Kick world & community! This is your chance to meet other entrepreneurial minds and get inspired by what’s possible. 👉 Learn more & Apply here: www.talentkick.ch/apply 🌟 Know someone who should be part of this? Tag them below👇 #TalentKick #StudentFounders #SwissInnovation #CoFounders #ScienceBasedStartups

👀 Still in the discovery phase and curious about entrepreneurship? Want to know more about Talent Kick and early-stage startup life? 🍂 This autumn, we're hosting an exclusive Talent Kick Immersion Day. What you’ll get: 🧠 Interactive workshops and inspiring panels to help you take action 📣 Insider tips and first-hand advice from founders and teams who joined Talent Kick 🤝 Real connections with like-minded talents and innovators from the Swiss startup ecosystem 🎤 Optional pitch opportunity to fast-track your application for TK-S26 before regular applicants This is a rare opportunity to: ✨ Experience what Talent Kick is all about 🤝 Connect with successful entrepreneurs, driven peers & program alumni 🚀 Learn more on how to navigate early-stage startup life If you're curious about entrepreneurship or actively exploring, looking for co-founders, or a driven entrepreneurial community, or interested in joining Talent Kick in the TK-S26 Batch or in the future, this is the perfect opportunity for you! 📩 Apply now to join Immersion Day! 👉 Learn more here: https://lnkd.in/e2mveRma 🌱 Already know our program? Want to be considered for early selection and pitch at Immersion Day? 👉 Fill out the Early Bird Application form: https://lnkd.in/eFm36S3r ⚠️ Spots are limited — and this is a rare opportunity, APPLY NOW! #TalentKick #ImmersionDay #StudentFounders #StartupLife #SwissStartups #TKCommunity