Machine Learning Acceleration for Fast Triggers

机器学习加速以实现快速触发

• 批准号: ST/W005565/1
• 负责人: Jim Brooke
• 金额: $ 3.14万
• 依托单位: University of Bristol
• 依托单位国家: 英国
• 项目类别: Research Grant
• 财政年份: 2021
• 资助国家: 英国
• 起止时间: 2021 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=ST%2FW005565%2F1
• 关键词: Machine; Learning; Acceleration; Fast; Triggers

Modern particle physics experiments generate vast amounts of data; far more than can possibly be stored. Experiments such as CMS and DUNE have built fast, complex, data processing systems, that can identify interesting events in the data, and save them for analysis. However these systems have limitations which can impact the precision of the measurements made by the experiment. Machine learning algorithms offer an exciting possibility to improve the performance of the data selection (or "trigger") systems. These algorithms are not typically fast enough for particle physics experiments, but a new generation of fast, programmable, processing devices may speed them up sufficiently to be useful. In this project we will evaluate the suitability of latest generation devices for these experiments, as well as developing machine learning algorithms which are fast enough, and performant enough, to improve the physics reach of CMS and DUNE.

现代粒子物理实验会产生大量数据；远远超出可能存储的更多。 CMS和Dune之类的实验已经构建了快速，复杂的数据处理系统，可以识别数据中有趣的事件，并保存它们进行分析。但是，这些系统具有可能影响实验测量的精度。机器学习算法为提高数据选择（或“触发”）系统的性能提供了令人兴奋的可能性。对于粒子物理实验，这些算法通常不够快，但是新一代的快速，可编程，处理设备可能会足够加快它们的速度。在这个项目中，我们将评估最新一代设备对这些实验的适用性，以及开发足够快且表现足够快的机器学习算法，以提高CMS和Dune的物理范围。