Abstract

Poster - Splinter EScience   (H 3005)

Unsupervised learning for agnostic knowledge discovery from simulations

Sebastian Trujillo Gomez, Kai Polsterer, Bernd Doser
Heidelberg Institute for Theoretical Studies (HITS)

Simulations are the best approximation to experimental laboratories in astrophysics and cosmology. However, the complexity and richness of their outputs severely limits the interpretability of their predictions. We describe a new approach to obtaining useful scientific insights agnostically from a broad range of simulations. The method can be used on today’s largest simulations and will be essential to solve the extreme data exploration and analysis challenges posed by the exascale era. Our concept is based on applying nonlinear dimensionality reduction to learn compact representations of the simulation in its intrinsic low-dimensional space. The simulation data is seamlessly projected onto this space for interactive inspection, visual interpretation, and quantitative analysis. We present a prototype using a Convolutional Autoencoder trained on simulated galaxies from IllustrisTNG to obtain a natural “Hubble tuning fork” similarity space that can be visualized interactively.