Abstract
Contributed Talk - Splinter EScience
Tuesday, 08 September 2026, 16:00 (MW-2235)
he challenges and rewards of developing useful discovery tools for modern astrophysics
Sebastian Trujillo Gomez
H-ITS
Machine learning–assisted analysis is now widespread in astrophysics, but its scope is still narrow, mostly focusing on individual tasks including regression, classification, and anomaly detection on large observational datasets. In contrast, the development of general visualization, exploration, modeling, and discovery tools has lagged behind. I will discuss the data analysis and modeling challenges posed by modern and future astrophysics scientific workflows, and the opportunities presented by the rapid development of powerful statistical techniques originating from the field of Machine Learning. As a prototypical example, I will describe the development process of our explorative discovery tools that aim to address these challenges, and the difficulties we face in building tools that are useful, interpretable, trustable, easy to use, and general and flexible enough to be widely adopted by the research community and beyond.