Workshop on Machine-Learning-Based Sampling in Lattice FieldTheory and Quantum Chemistry
October 25, 2024
The recent Bethe Forum, supported by TRA Matter and the CRC NuMeriQs, successfully brought together researchers from various fields facing similar challenges in their work.
Modern scientific research is increasingly interdisciplinary, with many fields now integrating artificial intelligence and machine learning. Despite tackling similar problems, these different sub-communities often struggle to "talk" to each other. This lack of communication presents a significant barrier to collaboration and sharing tools that could accelerate progress in multiple domains. For example, while the incorporation of physical symmetries into machine learning models has been widely studied in quantum chemistry, it’s only recently gaining traction in lattice quantum field theory.
To address this challenge, the workshop Machine-Learning-Based Sampling in Lattice Field Theory and Quantum Chemistry (add link to website) was initiated by Dr. Kim A. Nicoli and Prof. Lena Funcke from TRA Matter, along with Tom Froembgen (University of Bonn) and Dr. Shinichi Nakajima (BIFOLD). The event aimed to bring together academic and industry experts facing similar challenges that recently started to be tackled by machine learning methods.
A core focus of the workshop was on the Boltzmann distribution, an essential quantity in physical systems and critical for simulations and estimating physical observables. Recent advances in Generative AI have highlighted efficient protocols using machine learning for sampling from the target Boltzmann distribution of a given physical system. Both the lattice field theory and quantum chemistry communities have developed innovative tools for this purpose, and the forum provided an opportunity to explore how these tools might benefit each field.
The Bethe Forum was a great success, drawing a diverse audience including Master's and PhD students, postdoctoral researchers, faculty, and industry professionals. Notably, two technical staff members from cusp.ai, a pioneering startup focused on developing CO2-capturing materials through machine learning, participated in the workshop and presented some of their work. The forum featured a rich program with thirteen technical talks, a poster session, three panel discussions, a Bethe colloquium on Normalizing Flows for Lattice QCD by Dr.Gurtej Kanwar, and a TRA colloquium on Symmetries and Equivariance in AI for Science by Prof. Jan Gerken. Alongside the daily business, there was a lot of room for social gatherings, which was highly appreciated by the
participants. The organizing team thanks all the speakers for their insightful and accessible presentations, as well as the participants for fostering a relaxed, engaging environment that encouraged discussion and collaboration. We hope our efforts in bringing these communities togetherwill pave the way for future collaborations. Encouraged by the success of this first edition, the organizing team is already looking forward to planning the next edition for 2026.