Leading the Next Generation of Data-Driven Discoveries

点击翻译按钮获取中文摘要

Extracting accurate information from the modern data flood without sacrificing the ability to discover the unexpected is a core data challenge in many fields. A cross-disciplinary approach presents major opportunities to advance the state of the art in each. This project focuses on bringing together the state of the art in two fields: astrophysics, and humanitarian Earth observation. In astrophysics, the proliferation of data that will be available from upcoming large sky surveys presents significant challenges and opportunities. Accurate classification of this data (for example, whether a galaxy contains a buried, feeding supermassive black hole, or whether a new point of light in a galaxy is a supernova, and what kind) will enable major advancement in our understanding of how the Universe has evolved and will continue to evolve, and how galaxies such as the Milky Way form and grow within it. Following a natural disaster, responders need to know where roads are blocked, where buildings are damaged, and where survivors are sheltering. When satellites look down at the Earth instead of up at the heavens, they capture data that has some similar qualities but many complementary differences to astrophysical images. The driving need, however, is fundamentally the same. For instance, detecting the important changes before and after a storm or earthquake is similar to identifying the type of exploding star newly observed in a distant galaxy. In this example, astrophysical and humanitarian analysis have lessons to teach each other: Astrophysics has advanced methods of adjusting its algorithms to compensate for differences in observing conditions such as atmospheric turbulence, while humanitarian "Earth Observation" algorithms have tested promising methods of identifying critical changes based on only one before-and-after data point. There are many other examples of the potential for symbiosis, and this project will capitalise on this by working on both simultaneously in a cross-pollinating environment.Generally speaking, this research project will develop new tools for efficient, accurate data classification and labelling in these new data regimes where images and other data points arrive rapidly and are of varying quality and origin. Both fields make use of machine learning algorithms and combine machine classification with expert and high-quality crowd data labels. The project will test new techniques developed in each field on the other, using tool-specific expertise and combined domain knowledge to make new discoveries. By extracting insights from each regime and advancing their most effective tools, this work will enable us to understand the changing skies and the changing Earth in a way that provides real benefit (e.g. increased resilience, decreased recovery time, saving of lives) to distressed populations, maximising impact both near and far.The specific scientific topics this project will address cover some of the most pressing humanitarian needs across the globe and some of the most fundamental open questions about the Universe.