Speaker
Description
Throughout history, new observations of the night sky have led to new discoveries, often leading to paradigm shifts in astronomy. These have been made possible by observing new wavelength ranges; mapping smaller scales with an increasing spatial resolution; or peering deeper into the more distant universe; to name a few. Over the past decades, astronomy has been going through a data revolution, with numerous surveys mapping astronomical sources from radio to X-ray, and in some wavebands, also as a function of time. Although the discovery space has increased exponentially, trends and groups of interest may be more challenging to uncover due to the increased data complexity. However, such information-rich datasets offer a new opportunity: to use the data itself to form novel hypotheses, a core approach in the field of data science. This talk will focus on one particular type of complex astronomical datasets — multi-wavelength datasets fused from different large-scale surveys. I will start the talk by describing the major transformation astronomy has undergone in that respect, highlighting the challenges astronomers face when analyzing these heterogeneous and high-dimensional datasets. I will then describe how unsupervised machine learning algorithms can be used to reduce the dimensionality and visualize the data; identify simple trends or groups that extend throughout the different survey observables; and detect outliers that may represent completely new types of objects; and by that, expedite the discovery process.
