EAGER: ADAPT: AI-Enhanced Sampling for Lattice Field Theory and Beyond

EAGER：ADAPT：用于晶格场论及其他领域的人工智能增强采样

• 批准号: 2141336
• 负责人: Dries Sels
• 金额: $ 20.14万
• 依托单位: New York University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-09-01 至 2023-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2141336&HistoricalAwards=false
• 关键词: EAGER; ADAPT; AI; Enhanced; Sampling

This award will use Artificial Intelligence to greatly speed up theoretical calculations in elementary particle physics. Currently the detailed non-linear equations governing the interactions of elementary particles are not directly solvable, so a computational technique called Lattice Field Theory is used to approximately solve them at separate points in space (the lattice) and then combined to get an approximate result. The precision which can be obtained is only limited by computational time. Currently the largest High-Performance Computers in the world are kept busy with Lattice Field Theory equation solving. This work will apply a new technique called Normalizing Flows to greatly reduce the required compute time for these calculations. Normalizing flows are a class of deep generative models that can effectively model complex, high-dimensional distributions and have the potential to transform many areas of physics. These models are designed by composing many simple invertible neural network layers designed to efficiently compute the desired result. These flows can model the probability distribution of data from a real experiment and thus spend computational time only on the parts of the calculation that will dominate the required solution. This award will develop these normalizing flows for lattice field theory. The work will also enhance public outreach in the New York Public School system and maintain an active Artificial Intelligence social media presence.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

该奖项将利用人工智能大大加快基本粒子物理学的理论计算。目前，控制基本粒子相互作用的详细非线性方程无法直接求解，因此使用称为格场理论的计算技术在空间（晶格）中的不同点近似求解它们，然后组合得到近似结果。所能获得的精度仅受计算时间的限制。目前，世界上最大的高性能计算机都在忙碌着求解格场理论方程。这项工作将应用一种称为标准化流的新技术，以大大减少这些计算所需的计算时间。规范化流是一类深度生成模型，可以有效地模拟复杂的高维分布，并有可能改变许多物理领域。这些模型是通过组成许多简单的可逆神经网络层来设计的，这些神经网络层旨在有效地计算所需的结果。这些流可以对来自真实的实验的数据的概率分布进行建模，从而仅将计算时间花费在将主导所需解决方案的计算部分上。 该奖项将开发格场理论的这些规范化流程。这项工作还将加强纽约公立学校系统的公众宣传，并保持活跃的人工智能社交媒体存在。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的智力价值和更广泛的影响力审查标准进行评估来支持。

项目成果

Sampling QCD field configurations with gauge-equivariant flow models

使用仪表等变流量模型对 QCD 场配置进行采样

• DOI: 10.22323/1.430.0036
• 发表时间: 2023
• 期刊: Postcode Postbeanschriftungssysteme
• 作者: Abbott, Ryan;Albergo, Michael;Botev, Aleksandar;Boyda, Denis;Cranmer, Kyle;Hackett, Daniel;Kanwar, Gurtej;Matthews, Alexander;Racaniere, Sebastien;Razavi, Ali
• 通讯作者: Razavi, Ali

Gauge-equivariant flow models for sampling in lattice field theories with pseudofermions

用于在带有赝费米子的格场理论中采样的规范等变流模型

• DOI: 10.1103/physrevd.106.074506
• 发表时间: 2022
• 期刊: Physical Review D
• 影响因子: 5
• 作者: Abbott, Ryan;Albergo, Michael S.;Boyda, Denis;Cranmer, Kyle;Hackett, Daniel C.;Kanwar, Gurtej;Racanière, Sébastien;Rezende, Danilo J.;Romero-López, Fernando;Shanahan, Phiala E.
• 通讯作者: Shanahan, Phiala E.

Aspects of scaling and scalability for flow-based sampling of lattice QCD

• DOI: 10.1140/epja/s10050-023-01154-w
• 发表时间: 2022-11
• 期刊: The European Physical Journal A
• 作者: Ryan Abbott;M. S. Albergo;Aleksandar Botev;D. Boyda;Kyle Cranmer;D. Hackett;A. G. Matthews;S. Raca

Flow-based sampling in the lattice Schwinger model at criticality

临界点格子 Schwinger 模型中基于流的采样

• DOI: 10.1103/physrevd.106.014514
• 发表时间: 2022
• 期刊: Physical Review D
• 影响因子: 5
• 作者: Albergo, Michael S.;Boyda, Denis;Cranmer, Kyle;Hackett, Daniel C.;Kanwar, Gurtej;Racanière, Sébastien;Rezende, Danilo J.;Romero-López, Fernando;Shanahan, Phiala E.;Urban, Julian M.
• 通讯作者: Urban, Julian M.