Harnessing Machine Learning to Study the Life Cycle of Stars

像我们的太阳这样的恒星是在巨大的气体云中诞生的，这些气体云通常会同时产生数千颗恒星。当恒星形成时，它们将能量回馈到环境中(“反馈”)，并影响新生的气体云。这种反馈看起来像快速流动的气体，有时看起来像大气泡。数据很复杂，因此很难识别反馈。通常情况下，天文学家都是“亲眼”找到反馈的，这是主观和耗时的。计算机科学的一个新领域--机器学习--提供了另一种寻找反馈的方法。在机器学习中，计算机算法被训练成识别特征，就像人脑识别物体一样--比如猫、狗和汽车。研究小组将使用最先进的恒星形成模型来训练机器学习算法，以找到反馈。他们将把算法应用于望远镜观测，并与之前人类发现的反馈来源进行比较。研究人员将通过银河项目与公众分享这些模型，这是一个在线天文学项目，旨在培训人们在望远镜图像中识别气云的反馈。该项目还将培训学生的研究技巧，包括来自代表性不足群体的本科生。该提议解决了一个基本的恒星形成问题：在星际介质中，有多少质量和能量与恒星反馈有关？为了回答这个问题，PI和合作者将使用形成恒星的磁流体力学模拟来训练机器学习算法来识别和量化反馈，这些模拟的完整反馈信息是已知的。将产生尘埃和分子谱线“合成观测”，并将其与观测数据一起用作训练集。调查人员将把机器学习识别与之前视觉识别的反馈目录进行比较，包括公民科学银河项目的反馈目录，创建最新的人口普查，并向社区公开发布算法和数据。更广泛的影响目标是增加公众对银河系项目的参与，开发一个关于反馈的全球望远镜之旅，并培训学生，包括德克萨斯州天文学本科生研究经验不足学生(金牛座)暑期项目的本科生。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Stars like our Sun are born in large clouds of gas that usually produce thousands of stars at once. As the stars form, they add energy back into their environment ('feedback') and influence the birth gas cloud. This feedback appears as fast-moving gas, which sometimes looks like large bubbles. The data are complex, so identifying feedback is difficult. Typically, astronomers have found the feedback "by eye", which is subjective and time-consuming. A new field of computer science, machine learning, provides an alternative approach to find feedback. In machine learning, computer algorithms are trained to identify features in the same way the human brain recognizes objects - like cats, dogs and cars. The investigator's group will use state-of-the-art models of forming stars to train machine learning algorithms to find feedback. They will apply the algorithms to telescope observations and compare with the feedback sources previously found by humans. The investigator will share the models with the public through the Milky Way project, which is an online astronomy program that trains people to identify feedback in telescope images of gas clouds. The program will also train students in research techniques, including undergraduates from underrepresented groups. The proposal addresses a fundamental star formation question: How much mass and energy is associated with stellar feedback in the interstellar medium? To answer this question, the PI and collaborators will use magnetohydrodynamic simulations of forming stars, for which full feedback information is known, to train machine learning algorithms to identify and quantify feedback. Dust and molecular line 'synthetic observations' will be produced and used together with observational data as a training set. The investigators will compare the machine learning identifications to prior visually identified feedback catalogs, including those from the citizen-science Milky Way Project, create an updated census, and publicly release the algorithm and data to the community. The broader impact objectives are to increase public participation in the Milky Way Project, develop a WorldWide Telescope tour on feedback, and train students, including undergraduates in the Texas Astronomy Undergraduate Research experience for Under-Represented Students (TAURUS) summer program.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.