BEGIN:VCALENDAR VERSION:2.0 PRODID:-//CERN//INDICO//EN BEGIN:VEVENT SUMMARY:[ENCCS Webinar] Foundation Models for Atoms: Machine-Learned Inter atomic Potentials in Practice DTSTART:20260930T080000Z DTEND:20260930T113000Z DTSTAMP:20260901T052400Z UID:indico-event-502@events.hpc-portal.eu DESCRIPTION:About this webinar\nQuantum-mechanical methods such as density functional theory (DFT) are accurate but limited to small systems and sho rt timescales\, while classical force fields are fast but often not accura te or transferable enough. Machine-learned interatomic potentials (MLIPs) could break this trade-off between accuracy and scale\, and the field is n ow moving at remarkable speed. So-called universal or "foundation" models (e.g. MACE-MP\, UMA\, MatterSim\, Orb\, the DPA/OpenLAM series)\, pre-trai ned on tens to hundreds of millions of DFT calculations spanning the perio dic table\, approach DFT-level accuracy at a small fraction of the cost. T hey can be applied out-of-the-box ("zero-shot") to almost any chemistry\, then fine-tuned to a specific system with a small amount of targeted data\ , shortening the path from question to result for both academic and indust rial users.\nIn this webinar\, we will give a conceptual tour of this rapi dly evolving landscape and what it means for everyday computational resear ch on HPC systems. We will cover:\n\nwhat universal MLIPs are and how they differ from classical force fields and system-specific ML potentials\;\nt he practical pre-train → fine-tune workflow and when out-of-the-box use is (and is not) good enough\;\nthe new generation of batched\, GPU-acceler ated simulation engines (e.g. TorchSim\, kUPS) built for high-throughput M LIP simulation\;\nhow to choose and trust a model using open benchmarks\;\ nand close with a brief outlook on where the field is heading\, including ML making its way into electronic-structure theory itself\, with learned d ensity functionals\, and a short demo on European HPC resources. Throughou t\, we will point to open models\, datasets\, benchmarks\, and codes that participants can try on their own problems right after the webinar.\n\nWho is the webinar for\nThis webinar is intended for:\n\nResearchers and stud ents in computational materials science\, chemistry\, and condensed-matter physics who use DFT or ab initio MD and want to reach larger systems and longer timescales without giving up accuracy\nUsers of classical molecular dynamics (LAMMPS\, GROMACS\, ASE workflows) curious about upgrading to ML -based potentials\nIndustry R&D scientists and engineers exploring AI-acce lerated materials and molecular discovery\nHPC support staff and research software engineers who want an overview of the modern MLIP software stack and what it needs from GPU systems\nAnyone curious about how the foundatio n-model paradigm from language and vision AI is reshaping simulation in th e natural sciences\n\nNo prior experience with multiplet theory\, density functional theory (DFT)\, or many-body methods is required. Familiarity wi th basic concepts from solid-state or atomic physics is sufficient\; the r ole of data and HPC in modern electronic-structure studies will be introdu ced at a conceptual level.\nKey takeaways\nBy the end of this webinar\, pa rticipants will:\n\nUnderstand what universal machine-learned interatomic potentials are and why they deliver near-DFT accuracy at a fraction of the cost\nKnow the practical pre-train → fine-tune workflow: when an off-th e-shelf foundation model is sufficient\, when and how to fine-tune it with a small targeted dataset\nGet an overview of the modern MLIP software sta ck\, from ASE/LAMMPS integrations to GPU-native\, batched engines such as TorchSim\, and what it takes to run it efficiently on EuroHPC systems\nBe able to critically select and validate a model using open benchmarks and u nderstand why no single model wins on every axis\nGain an outlook on emerg ing directions: long-range electrostatics\, MLIPs that predict polarisatio n and spectra\, the use of machine learning within DFT itself\, and early work on AI agents that help orchestrate computational workloads\n\nSpeaker and moderator\n\nKarim Elgammal\nYonglei Wang/Wei Li\n\nFor any questions contact us at training@enccs.se\nMore events & contact\nCheck out more up coming events from ENCCS and our European network at HERE\, as well as ava ilable ENCCS lesson materials\, suitable also for self-learning.\nFor ques tions regarding this workshop or general questions about ENNCS training ev ents\, please contact training@enccs.se\nSchedules can change!\nTo ensure that everyone has the opportunity to participate\, we kindly request that you let us know as soon as possible if you are unable to attend an event a fter registering.\nPlease send us an email at training@enccs.se to cancel your attendance.\nWe understand things can change\, but repeated cancellat ions without notice may unfortunately result in your name being removed fr om future event registration lists.\nRegulations\nDue to EuroCC2 regulatio ns\, we CAN NOT ACCEPT generic or private email addresses. Please use your official university or company email address for registration.\nThis trai ning is intended for users established in the European Union or a country associated with Horizon 2020. You can read more about the countries associ ated with Horizon2020 HERE.\n\nhttps://events.hpc-portal.eu/event/502/ LOCATION:Online URL:https://events.hpc-portal.eu/event/502/ END:VEVENT END:VCALENDAR