Accelerating Discovery in Collider Physics with Foundation Models | Vinicius Mikuni скачать в хорошем качестве

Accelerating Discovery in Collider Physics with Foundation Models | Vinicius Mikuni

FirstPrinciples Talks presents 'Accelerating Discovery in Collider Physics with Foundational Models' To learn more about FirstPrinciples and the AI Physicist: https://www.firstprinciples.org/ Speaker: Vinicius Mikuni Foundation Models are neural networks designed to solve many tasks at once, and large language models have already reshaped daily life. But what does the “foundation model revolution” mean for scientific discovery? In this talk, Vinicius Mikuni introduces OmniLearn, a foundation model built for collider physics, designed to adapt across tasks and datasets. OmniLearn points toward a future where foundation models become practical tools for high-energy physics (not just proof-of-concept demonstrations). Vinicius highlights how foundation models can help address three major challenges in collider physics: Reduce computing costs when developing reconstruction algorithms Enable full uncertainty quantification for high-dimensional measurements Support model-agnostic searches for new physics, using low-level detector inputs Each of these problems has long been limited by major computational and methodological bottlenecks, limiting the real-world science impact of deep learning in particle physics. By tackling them directly, foundation models like OmniLearn could push jet and collider ML beyond narrow benchmarks and into the everyday toolkit of physics practitioners. This talk is for anyone interested in AI for science, collider physics, machine learning for HEP, uncertainty quantification, and searches for new physics. About the speaker Vinicius Mikuni is a Machine Learning Postdoctoral Fellow at Berkeley Lab and an incoming Associate Professor at the Kobayashi-Maskawa Institute in Nagoya, Japan. His research lies at the intersection of machine learning and fundamental science, where he develops algorithms to tackle core challenges in physics. His recent work includes fast simulation frameworks for fluid dynamics, collider physics, nuclear physics, and astrophysics using diffusion-based generative models; innovative methods for solving inverse problems in collider and neutrino physics; and leveraging pre-trained models to accelerate discoveries in particle physics. He earned his PhD in 2021 from the University of Zurich, where he measured the production cross-section of top quark pairs in association with b quarks and led searches for new physics in signatures involving third-generation fermions using CMS data. Subscribe for more talks on AI-enabled scientific discovery. #ColliderPhysics #particlephysics #machinelearning #foundationmodels #aiinscience #deeplearning #HEP #newphysics #ai