Investigations into new machine learning techniques for gravitational wave astronomy

引力波天文学新机器学习技术的研究

• 批准号: 2748216
• 金额: --
• 依托单位: University of Glasgow
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2022
• 资助国家: 英国
• 起止时间: 2022 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2748216
• 关键词: Investigations; into; new; machine; learning

The field of gravitational wave astronomy is currently benefiting from the rapid growth of machine learning applications to problems such as classification (detection or model selection) and regression (parameter estimation). So far, only these areas of gravitational wave data analysis have been investigated since there is a direct relation between them and the most common and well established machine learning processes. In this project we propose the investigation of a new (known) form of machine learning known as Physics Informed Neural Networks (PINNs) in which machine learning is used to obtain accurate models of functions that are particular solutions to physically informed problems. For example, such problems include gravitational waveforms modelled by Einsteins General Relativity, solutions to the Tolman Oppenheimer Volkoff Equation for neutron stars, solving geodesic paths for particles in a general relativistic spacetime, and solving the spacetime metric for arbitrary mass-energy distributions. The student will investigate the potential for PINNs amongst these and other problems in the gravitational wave field.

引力波天文学的领域目前正在从机器学习应用的快速增长中受益于分类（检测或模型选择）和回归（参数估计）等问题。到目前为止，由于它们与最常见且最合适的机器学习过程之间存在直接关系，因此仅研究了重力波数据分析的这些领域。在这个项目中，我们建议对一种新的（已知）形式的机器学习形式进行调查，称为物理学知情的神经网络（PINN），其中使用机器学习来获得准确的功能模型，这些功能是特定解决方案的特定解决方案，以解决物理知情问题。例如，此类问题包括以Einsteins的一般相对性建模的引力波形，对中子恒星的Tolman Oppenheimer Volkoff方程的解决方案，求解了一般相对论的时空中粒子的地质路径，并求解了Quyary Mass-Mass-Mass-Mass-Mass-Mass-emenergy分布的空间时间。学生将研究引力波场的这些和其他问题之间的PINN的潜力。