Transcending the Limits of Theoretical Astrophysics with Deep Learning
Abstract:
Astronomy has undergone a significant transformation in recent times, as the acquisition of copious amounts of data through increasingly powerful instruments has opened up a plethora of new avenues for exploration. However, this blessing is not without its own set of challenges, as astronomical phenomenology can often be intricate and inherently high-dimensional in nature, encompassing meticulous imaging, spectra, and time series with utmost precision. In this realm, conventional astrostatistical methods falter. One of the fundamental features of deep learning is that it overcomes the curse of dimensionality, thus enabling a faithful representation of complex phenomenology. I will elaborate on deep learning approaches for characterizing complex astronomical systems, while also delving into the theoretical foundation of deep learning, including its intertwined relation with symmetry and physics. Moreover, I will explore a myriad of applications in asteroseismology, weak stellar spectroscopy, weak lensing, reionization, galactic dynamics, galaxy evolution, and quasars, all of which contribute to expanding the horizon of Bayesian statistics and theoretical astrophysics through the lens of deep learning.
