Machine Learning for High-Energy Astronomy Surveys

我研究的目的是通过实现一个系统来从根本上改进现有的源检测算法，该系统可以在所有时间尺度上搜索天空中的所有点，以寻找重要的排放物。这将涉及寻找“更智能”的调查方法，并应用新的数据科学和机器学习技术，特别是瞬态检测。我使用监督式深度学习方法，如卷积神经网络，来检测并最终分类持久性和瞬态X射线源。为了寻找其中一个源何时处于爆发状态，我使用了无监督的方法，例如对时间序列序列进行k均值聚类，这是对时间数据进行异常检测。

The aim of my research is to radically improve existing source detection algorithms by implementing a system that searches all points on the sky on all timescales for significant emission. This will involve looking at 'smarter' ways to do surveys, and applying new data science and machine learning techniques, especially for transient detection.I use supervised deep learning methods such as convolutional neural networks to detect and eventually classify both persistent and transient X-ray sources. To look for when one of these sources is in outburst I use unsupervised methods such as k-means clustering of time series subsequences which is using anomaly detection on the temporal data.