RAISE: Deep Gravitational Wave Exploration, Instrumental Insights and Noise Removal Through Machine Learning

这是由综合活动办公室、计算机与信息科学与工程理事会（CISE）计算与通信部（CCF）的信号处理系统项目、数学与物理科学理事会（MPS）的多学科活动办公室以及MPS物理部的引力物理项目支持的RAISE奖。最近发现的黑洞碰撞产生的引力波揭示了一个研究宇宙的新时代。未来几年，黑洞合并以及其他引力波来源（包括恒星爆炸）的探测将会激增。对于每一个发现，都会有许多微弱的引力波信号被埋在探测器的噪音中，很难被发现。引力波探测器是一个极其复杂的系统，噪声源会以无数种独立的方式干扰记录的数据，偶尔会产生奇怪的数据伪像，很难与引力波区分开来。机器学习非常适合理解这种复杂性，并将数据从噪音中分离出来，拓宽我们探测引力波的视野。pi将设计机器学习技术来理解LIGO的40万个辅助数据通道，并识别探测器行为模式，以便在高度非线性和非高斯背景噪声中识别宇宙信号。pi将研究和使用最优策略，包括稀疏回归和鲁棒主成分分析，从天体物理信号中区分探测器或环境人工影响，以发现引力波，否则这些引力波可能仍然是不可见的。

This is a RAISE Award supported by the Office of Integrative Activities, the Signal Processing Systems program of the Division of Computing and Communications (CCF) of the Computer & Information Science & Engineering Directorate (CISE), the Office of Multidisciplinary Activities of the Mathematical and Physical Sciences Directorate (MPS) and the Gravitational Physics program of Physics Division in MPS. The recent discovery of gravitational waves from colliding black unveiled a new era of broad opportunities for studying the Cosmos. The coming years will bring about the proliferation of detections of black hole mergers as well as other sources of gravitational waves, including stellar explosions. For every discovery, there will be numerous weak gravitational wave signals buried in the detector noise that will be difficult to unearth. Gravitational-wave detectors are incredibly complex systems, where there are myriads of independent ways noise sources can interfere with the recorded data, occasionally producing curious data artifacts that are difficult to distinguish from gravitational waves. Machine learning is uniquely suited to make sense of this complexity, and disentangle data from the noise to broaden our horizon to detecting gravitational waves.The PIs will design machine-learning techniques to make sense of LIGO's 400,000 auxiliary data channels and identify patterns in detector behavior to enable the identification of cosmic signals in the midst of highly non-linear and non-Gaussian background noise. The PIs will research and use optimal strategies, including sparse regression and robust principal component analysis, to distinguish detector or environmental artifacts from astrophysical signals to discover gravitational waves that otherwise could have remained invisible.