SCMP100 SciNet User Group Meeting (Sept. 2018)

George Stein (Dept. of Astronomy-UofT, CITA) Machine learning cosmic structure formation + pizza Abstract: In modern astrophysics and cosmology, accurate simulations of the large scale structure of the universe are necessary. Usually, this is accomplished by so called N-body simulations, which calculate the full gravitational collapse of a region of the universe over its 14 billion year history. Instead of calculating this costly gravitational evolution, we trained a three-dimensional deep Convolutional Neural Network (CNN) to identify dark matter proto-haloes directly from the cosmological initial conditions, and showed that a CNN of this type can be a viable alternative in some cases. In this talk I will discuss current cosmological simulations and the invasion of machine learning techniques, with a focus on our work. For more information see https://arxiv.org/abs/1805.04537 Please register for pizza purposes!