Machine learning meets physics: A two-way street

• 成果类型: Article
• 署名作者: Levine, Herbert; Tu, Yuhai
• 署名单位: Northeastern University; International Business Machines (IBM); IBM USA
• 刊物名称: PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
• ISSN/ISSBN: 0027-14177
• DOI: 10.1073/pnas.2403580121
• 发表日期: 2024-07-02
• 关键词: neural-networks; protein; landscape; models; go

摘要

This article introduces a special issue on the interaction and ongoing research in physics. The first half of the papers in this issue deals with the question, what can machine learning do for physics? The second part asks the reverse, what can physics do for machine learning? As we will see, both of these directions are being vigorously pursued. Physics is, of course, a very broad discipline, and almost every part of it has been exploring the possible use of machine learning (ML). We obviously cannot cover all of these developments systematically. Instead, we will present various examples, and try to propose some tentative general insights. Given the tremendous buzz of activity, we are sure that our perspective will need to be constantly revised in the light of accumulating experience. Nevertheless, we proceed.