争取每周一更，精选一个本周国际计算机学会科技新报的最有趣研究成果～
2018年1月30日
深度学习驱动的引力波探测
关键词：深度过滤器模型
伊利诺伊大学香槟分校NCSA的研究者开发了基于GPU加速的深度学习框架来检查和识别引力波。深度学习采用了结合相对论数值模拟和LIGO实测数据的过滤器，以及端到端的时间序列信号处理方法。与传统引力波检测算法相比，深度过滤器在相似的敏感性前提下，达到更低的错误率和更加高效抗噪。NCSA的蓝水超级计算机（Blue Waters supercomputer）实现了速度快于LIGO数据实时采集的引力波信号处理。更值得一提的是，研究者设计了一套可视化演示方法，更好的展现深度过滤器模型在检测识别引力波事件过程中的神经元活动。研究斩获2017年超级计算会议的ACM学生研究竞赛冠军（ACM Student Research Competition at the SuperComputing 2017 (SC17) conference in Denver, CO）。
Researchers at the University of Illinois at Urbana-Champaign's National Center for Supercomputing Applications (NCSA) have developed a method for using graphics processing unit-accelerated deep learning for detecting and characterizing gravitational waves. The researchers used deep learning algorithms, numerical relativity simulations of black hole mergers, and data from the LIGO Open Science Center to produce Deep Filtering, an end-to-end time-series signal processing method. Deep Filtering achieves similar sensitivities and lower error rates when compared to conventional gravitational wave detection algorithms, and is more computationally efficient and resilient to noise anomalies. The method enables faster-than-real-time processing of gravitational waves in LIGO's raw data, thanks to NCSA's Blue Waters supercomputer. In addition, the researchers created a demonstration to visualize the architecture of Deep Filtering, and gained insights into its neuronal activity during the detection and characterization of real gravitational wave events. Their research won first place in the ACM Student Research Competition at the SuperComputing 2017 (SC17) conference in Denver, CO last November.
