New Methods for Observational Cosmology and Galactic Archaeology

我们能否利用机器智能来发现庞大天文数据集的复杂性和相互关系，从而理解宇宙结构和恒星系统？从天空调查中产生的大量数据需要有效的技术来将原始调查数据提炼成物理见解。本研究计划旨在开发新的方法，以最大限度地从多个天空调查中联合提取物理量。我们将利用可扩展的机器智能和统计技术来分析多波长的天体物理数据，重点是银河考古学和宇宙学。通过收集数百万颗恒星的光谱，从公共天空调查和最先进的模拟中，我们将实现智能机器来绘制我们的宇宙。该系统将学习多波长数据的复杂性，特别强调传统上难以解释的仪器和大气污染物。由此产生的系统将以概率输出预测恒星的一致温度、金属丰度、di和化学丰度。我们将通过纳入来自广域成像调查的信息来改进我们的系统，从而提供预测恒星质量、年龄和距离的额外能力。我们的目标是建立一个银河系学习机器来研究恒星系统的化学演化，以及空前水平的元素核合成。与此同时，我们将为银河系外的资源设计智能系统。欧几里得联盟是此类应用的一个主要例子：地面和太空望远镜收集数百万张河外天空的光学和红外波长图像和光谱，以探测宇宙的黑暗部分。我们将进化我们的银河系机器，从星系像素、空间分布和光谱中学习，并进行模拟，增加预测河外红移和约束宇宙学模型的能力。该计划非常注重培养研究生在这个大天文数据的黄金时代创造和开发新颖的机器学习技术。该项目支持的加拿大青年研究人员将获得强大的学术研究和工业数据科学技能。

Can we use machine intelligence to discover complexities and relationships within huge astronomy data sets to understand cosmic structure and stellar systems? The enormous quantities of data generated from sky surveys require efficient techniques to distill raw survey data into physical insights. This research proposal aims at developing novel methods to maximize the extraction of physical quantities jointly from multiple sky surveys. We will take advantage of scalable machine intelligence and statistical techniques to analyze multi-wavelength astrophysical data with a focus on galactic archaeology and cosmology. By assembling spectra of millions of stars, from public sky surveys and state-of-the-art simulations, we will implement smart machines to map our universe. The systems will learn the complexities of the multi-wavelength data with a special emphasis on instrumental and atmospheric contaminants, traditionally difficult to account for. The resulting systems will predict consistent temperatures, metallicities, di and chemical abundances of the stars with probabilistic outputs. We will evolve our systems by including information from wide field imaging surveys, giving the extra capacity to predict mass, age and distances of the stars. Our goal is to build a Milky Way learning machine to study the chemical evolution of stellar systems, and the nucleosynthesis of elements at an unprecedented level. In parallel we will design smart systems for extragalactic sources. The Euclid consortium is a prime example for such an application: ground and space telescopes collecting millions of images and spectra of the extragalactic sky at optical and infrared wavelengths, to probe the dark sector of the universe. We will evolve our Milky Way machine to learn from galaxy pixels, spatial distributions and spectra together with simulations, adding a capacity to predict extragalactic redshifts and constrain cosmological models. The program has a strong focus on training graduate students to create and develop novel machine learning techniques in this Golden Era of Big Astronomy Data. Young researchers in Canada supported by this program will acquire strong skills for both academic research and data science in industry.