Reposted by @claudius-krause.bsky.socialmla2s-oeaw.bsky.social @mla2s-oeaw.bsky.social · 16hAI 🤖 meets science 🔬 at the MLA2S Networking Seminar #11. Join us for an afternoon of short talks, exchange and informal networking. 📅 Oct 15, 1 PM 📍 PSK Building, Seminar Rooms 2 & 3 More info & registration👉 bit.ly/4zyTxAs 022
claudius-krause.bsky.social @claudius-krause.bsky.social · 12/10/2026Everything I get to review that says SI instead of AI will be rejected for not properly using scientific nomenclature 140
claudius-krause.bsky.social @claudius-krause.bsky.social · 08/10/2026Like observed here? arxiv.org/abs/2609.310...arxiv.orgConfident, Not Wiser: The Dunning-Kruger Effect in Human-AI Interaction 000
Reposted by @claudius-krause.bsky.socialAustrian Scientific Computing (ASC) @asc.ac.at · 06/10/2026A new preprint by ASC users looks at how much compute it takes to generate one billion particle-collision events for the #LargeHadronCollider. The team combines #GPU-based calculations with normalizing flows, cutting generation time from 47 days to 14 hours on 4 H100 GPUs of MUSICA. 011
claudius-krause.bsky.social @claudius-krause.bsky.social · 02/10/2026Is this already being enforced by arXiv? I couldn't find a statement by them on their blog or in their code of conduct... 100
claudius-krause.bsky.social @claudius-krause.bsky.social · 01/10/2026New arXiv rule: max 2 submissions/author/month 010
claudius-krause.bsky.social @claudius-krause.bsky.social · 30/09/2026Indeed, a very lovely essay! 000
Reposted by @claudius-krause.bsky.socialCERN @cern.bsky.social · 29/09/2026Happy birthday CERN! 🎉 #OnThisDay 72 years ago, CERN officially came into being, bringing together 12 founding Member States with a shared vision for science and international collaboration. Discover the story behind 72 years of science: home.cern/about/who-we... 10484132
claudius-krause.bsky.social @claudius-krause.bsky.social · 26/09/2026I wasn't sure what to do with an email saying "you have been nominated for Full Membership in Sigma Xi, The Scientific Research Honor Society" I guess ignoring it was a good decision 190
Reposted by @claudius-krause.bsky.socialAnthropic {bot} @anthropicai.xmirror.bot · 25/09/2026New on the Science Blog: Yes, Claude can do Nine Loops.anthropic.comClaude gives us N=4 super Yang-Mills to nine loopsClaude gives us N=4 super Yang-Mills to nine loops 1184
Reposted by @claudius-krause.bsky.socialMarietta-Blau-Institut für Teilchenphysik @mbi-oeaw.bsky.social · 24/09/2026We are hiring! 💥PostDoc - Working on AI Applications 💥 full-time 💥 for a period of two years 🫵 Apply now! www.oeaw.ac.at/jobs?jh=uy5i...oeaw.ac.atOffene Stellen an der ÖAWAktuelle Stellenausschreibungen aus den Instituten und der zentralen Verwaltung der ÖAW 021
claudius-krause.bsky.social @claudius-krause.bsky.social · 23/09/2026Not TCS, but other (non math) fields: The Austrian Academy of Sciences (ÖAW) recently asked the same question. There are a few replies online, including my take on particle physics www.oeaw.ac.at/preisfrage/p...oeaw.ac.atEssays 000
Reposted by @claudius-krause.bsky.socialMarietta-Blau-Institut für Teilchenphysik @mbi-oeaw.bsky.social · 23/09/2026Der Kran war übrigens 80 m (!) hoch! 031
Reposted by @claudius-krause.bsky.socialMarietta-Blau-Institut für Teilchenphysik @mbi-oeaw.bsky.social · 22/09/2026Nach monatelanger Planung war es gestern endlich so weit: Die CNC-Fräse aus der Nikolsdorfergasse im 5. Bezirk wurde in die PSK im 1. Bezirk übersiedelt. Davor gab es jedoch viele Zwischenschritte, wie man den Bildern entnehmen kann! Großartige Leistung an das gesamte Team! 🙌💪 041
claudius-krause.bsky.social @claudius-krause.bsky.social · 21/09/2026And thanks again to all our sponsors: @univie-physik.bsky.social @univie.ac.at @wien.gv.at #NextGenerationTriggers project (#NGT) at @cern.bsky.social @oeaw.bsky.social and @mbi-oeaw.bsky.social ! Thank you very much! 020
claudius-krause.bsky.social @claudius-krause.bsky.social · 21/09/2026If you want to check out the talks again, you can find them here: indico.global/event/15240/... #ML4Jetsindico.globalML4Jets 2026Over the past years, High Energy Physics (HEP) has made significant strides by developing and applying machine learning (ML) and artificial intelligence (AI) approaches. These advancements have greatl... 110
claudius-krause.bsky.social @claudius-krause.bsky.social · 18/09/2026And we're done. Thanks to everyone for attending and making #ML4Jets such a great conference! See you next year in #Nagoya 132
claudius-krause.bsky.social @claudius-krause.bsky.social · 18/09/2026Now up at #ML4Jets: Manuel Szewc, talking about "Machine Learning for Hadronization" 000
claudius-krause.bsky.social @claudius-krause.bsky.social · 18/09/2026Today is the last day of #ML4Jets 2026 in #Vienna We kick it off with Felix Weiglhofer, talking about "GPU-Accelerated Online Reconstruction in ALICE" 100
claudius-krause.bsky.social @claudius-krause.bsky.social · 17/09/2026And Sioni Summers with "Edge ML in Particle Physics: Bringing Inference to the Detector" #ML4Jets 000
claudius-krause.bsky.social @claudius-krause.bsky.social · 17/09/2026Day 4 of #ML4Jets 2026 in #Vienna! Today's first topic is "Anomaly Detection: From Methods to Applications" by Marie Hein 100
claudius-krause.bsky.social @claudius-krause.bsky.social · 16/09/2026Next up, Eric Moreno on "From Prompts to Publications: Agentic AI for Collider Analysis" #ML4Jets 000
claudius-krause.bsky.social @claudius-krause.bsky.social · 16/09/2026And we're kicking off day 3 of #ML4Jets in #Vienna! Anna Hallin starts with "Foundation Models for High Energy Physics" 101
claudius-krause.bsky.social @claudius-krause.bsky.social · 15/09/2026But that way you are showing that you personally write, not a LLM/agent for you ;-) 110
claudius-krause.bsky.social @claudius-krause.bsky.social · 15/09/2026And Jay Sandesara with "Neural Simulation-Based Inference at the LHC" #ML4Jets 000
claudius-krause.bsky.social @claudius-krause.bsky.social · 15/09/2026Next up, Michal Mazurek giving an "Overview of the Application of Generative Models to Fast Simulation in Large HEP Experiments" #ML4Jets 100
claudius-krause.bsky.social @claudius-krause.bsky.social · 15/09/2026Welcome to Day 2 of #ML4Jets 2026 in Vienna! Today we start with Theo Heimel: "Machine Learning for LHC Event Generation" 110
claudius-krause.bsky.social @claudius-krause.bsky.social · 14/09/2026First plenaries at #ML4Jets 2026 in #Vienna: Nicole Hartman with "ML for reconstruction: A journey through deep learning’s impact for LHC experiments" And Benoît Assi with "Machine Learning Before the Collision: AI for Theory and Phenomenology" 030
claudius-krause.bsky.social @claudius-krause.bsky.social · 14/09/2026First, with some warm welcomes. Here, Jochen Schieck, director of @mbi-oeaw.bsky.social 000
claudius-krause.bsky.social @claudius-krause.bsky.social · 14/09/2026And we start! #ML4Jets 2026 in Vienna just kicked off! indico.global/event/15240/...indico.globalML4Jets 2026Over the past years, High Energy Physics (HEP) has made significant strides by developing and applying machine learning (ML) and artificial intelligence (AI) approaches. These advancements have greatl... 110
Reposted by @claudius-krause.bsky.socialAnima Anandkumar @anima-anandkumar.bsky.social · 08/09/2026We have found stable singularity on 3D Euler! tensorlab.cms.caltech.edu/users/anima/... We have been working on this problem for much of this year, and we just saw the announcement by Tristan on Euler with forcing. In contrast, we consider without forcing and use a PINN formulation. @caltech.eduanima-ai.orgStable Singularity of the Euler Equations on R^3 without forcing | Anima on AI 031
Reposted by @claudius-krause.bsky.socialOpenAI @openai-m.extwitter.link · 08/09/2026We congratulate Levent Alpöge and Tristan Buckmaster on their remarkable mathematical work. We (the researchers and the agents) did not see any of their work through any means until they released it publicly — in particular, no specific user data was accessed in... 🔗 twitter.com/i/status/209737... 052
Reposted by @claudius-krause.bsky.socialEthan Mollick @emollick.bsky.social · 08/09/2026This is a VERY big one. (And yes, the fights over academic credit and what happened in the race for the proof needs to be resolved, but it is still appears that this is a big one, if true.) openai.com/index/navier...openai.comOn the Navier–Stokes Millennium Prize ProblemWe’re sharing an AI-generated solution to the Navier–Stokes Millennium Prize Problem, including a writeup and a formal proof in Lean. 619731
Reposted by @claudius-krause.bsky.socialLukas Heinrich @lukasheinrich.com · 03/09/2026Postdoc Fellowship 🚨 (Sept 30) We're hiring a postdoc on our ORIGINS Data Science Lab fellowship. The position allows researchers in particle physics to pursue their independent data-science / AI-related research vision in the context of our Excellence Cluster. inspirehep.net/jobs/3199251inspirehep.netINSPIRE 044
Reposted by @claudius-krause.bsky.socialÖsterreichische Akademie der Wissenschaften @oeaw.bsky.social · 19/08/2026Wie viel wissen wir wirklich über das Universum? ✨ Am 10. September spricht CERN-Generaldirektor Mark Thomson auf Einladung des @mbi-oeaw.bsky.social an der #ÖAW über Higgs, den LHC und die Zukunft der Teilchenphysik. 🗓️10.9., 18:00 Eintritt frei Veranstaltung auf Englisch 074
Reposted by @claudius-krause.bsky.socialTeX Users Group Official Account @texusersgroup.bsky.social · 31/08/2026From #CTAN: Lawrence Lee submitted the INSPIRE-HEP package. Version: 0.2 License: lppl1.3c Summary description: An INSPIREHEP × LaTeX interface for automatic reference management ctan.org/pkg/inspirehep #TeXLaTeX @inspirehep.netctan.orgCTAN: Package INSPIRE-HEP 043
claudius-krause.bsky.social @claudius-krause.bsky.social · 01/09/2026There's a lot more to generative AI in particle physics than language models, just saying 010
Reposted by @claudius-krause.bsky.socialThe Linux Foundation @linuxfoundation.org · 25/08/2026"I'm doing a (free) operating system (just a hobby...)" On Aug 25, 1991, a 21-year-old student shared his side project to a Usenet newsgroup. 35 years later, Linux powers the global cloud, space exploration, and billions of devices. Happy 35th Birthday, Linux! 🐧✨ bit.ly/4gkst16 #Linux35 5466117
claudius-krause.bsky.social @claudius-krause.bsky.social · 24/08/2026Hey @asc.ac.at, our paper hit the arXiv today: bsky.app/profile/clau... and arxiv.org/abs/2608.21338arxiv.orgEfficient Event Generation for High-Multiplicity LHC Processes: An End-to-End GPU Workflow with Normalizing FlowsProducing very large unweighted event samples for high-multiplicity processes is limited by expensive matrix-element evaluations and low unweighting efficiencies. We present the first end-to-end GPU-r... 000
claudius-krause.bsky.social @claudius-krause.bsky.social · 24/08/2026Coming back to the original question: "Can I generate 1B events?" Yes! For pp -> e⁺e⁻ + 4jets this now takes 14 hours instead of 47 days on the 4 H100 GPUs of MUSICA (@asc.ac.at)! For more details, numbers, and other processes, check our preprint: arxiv.org/abs/2608.21338arxiv.orgEfficient Event Generation for High-Multiplicity LHC Processes: An End-to-End GPU Workflow with Normalizing FlowsProducing very large unweighted event samples for high-multiplicity processes is limited by expensive matrix-element evaluations and low unweighting efficiencies. We present the first end-to-end GPU-r... 020
claudius-krause.bsky.social @claudius-krause.bsky.social · 24/08/2026"Grouped flows" consider subprocesses with similar particle content in initial and final state with a single normalizing flow that is conditioned on the subprocess number. Now pp -> e⁺e⁻ + 4jets requires only 13 Normalizing Flows. The results are similar: speed-ups of up to 2 orders of magnitude 110
claudius-krause.bsky.social @claudius-krause.bsky.social · 24/08/2026"Subprocess-specific flows" means one flow per subprocess. For the example process pp -> e⁺e⁻ + 4jets, that's 134 normalizing flows! Nevertheless, we see speed improvements over the default Pepper of up to 2 orders of magnitude! 100
claudius-krause.bsky.social @claudius-krause.bsky.social · 24/08/20262) We use Normalizing Flows to generate the random numbers. In particular, we study 2 different setups to combine the partonic subprocesses into the final event. 100
claudius-krause.bsky.social @claudius-krause.bsky.social · 24/08/2026In our preprint, we combine two previous ideas to speed up the event generation process: 1) Matrix elements are computed on the GPU. With the event generator Pepper, we eliminate the time-consuming back and forth betwenn CPU and GPU that we faced in previous publications. 100
claudius-krause.bsky.social @claudius-krause.bsky.social · 24/08/2026For many particles in the final state, this unweighting step can mean that 99.99% of all sampled events are thrown away, making the generation of unweighted events a time consuming enterprise. 100
claudius-krause.bsky.social @claudius-krause.bsky.social · 24/08/2026The second part concerns the sampling of the momenta according to that probability. We use proposal functions for this that are close to the true distribution. With an accept/reject algorithm in the end, we correct all "wrong" sampling by throwing away what doesn't match the desired distribution. 100
claudius-krause.bsky.social @claudius-krause.bsky.social · 24/08/2026High multiplicity events face two bottlenecks. 1) Slow matrix element evaluation and 2) low unweighting efficiency. The first part concerns the calculation of the differential cross section - how likely an event with a given momentum configuration is. Up until recently, this was done on a CPU. 110
claudius-krause.bsky.social @claudius-krause.bsky.social · 24/08/2026Can you simulate a billion particle collisions in a reasonable time? How much do you gain when entirely on a GPU? We looked at these questions in our preprint "Efficient Event Generation for High-Multiplicity LHC Processes: An End-to-End GPU Workflow with Normalizing Flows" arxiv.org/abs/2608.21338 110
