Cosmology and galaxy astrophysics with simulations and machine learning 2027

Description

The next few years mark a watershed moment for extragalactic astrophysics and cosmology. Surveys such as DESI, LSST, Euclid, Roman, and SKA are delivering — or are on the verge of delivering — datasets of unprecedented scale, with the potential to transform our understanding of galaxy evolution and fundamental cosmological parameters. These surveys will enable detailed, statistically robust studies of how galaxies form, evolve, and interact with their environments across cosmic time, linking small-scale baryonic physics to large-scale structure in ways that have not previously been possible. Meeting this moment demands a new synthesis of data, cosmological simulations, machine learning, and simulation-based inference, and a community that knows how to wield them together.

Substantial advances have been made on each front. Large N-body suites and hydrodynamic simulations now provide the computational scaffolding to model the universe across a wide range of scales and physical prescriptions, and various simulation projects are enabling machine learning models that simultaneously maximize cosmological information extraction and marginalize over astrophysical uncertainties. Neural network-based field-level inference has begun to outperform traditional power-spectrum analyses, and simulation-based inference (SBI) methods are maturing rapidly, enabling forward-modeling frameworks that jointly constrain cosmology and galaxy evolution by directly connecting observed galaxy distributions to the nonlinear physical processes that shape them.

Yet significant challenges remain. Chief among them are robustness and uncertainty quantification: how do we ensure that models trained on simulations perform reliably on real survey data, with all its instrumental systematics and baryonic complexity? How do we connect the inference machinery built on simulations to the actual outputs of ongoing surveys? And as AI agents become increasingly capable, how should they be integrated into research workflows without sacrificing interpretability? Looking ahead five years, the way research in galaxy evolution and cosmology is conducted will look fundamentally different — charting that path thoughtfully, and identifying the gaps that must be closed to get there, is an urgent collective task.

The goal of this workshop is to bring together a diverse community spanning simulations, machine learning, and observational cosmology to take stock of the rapidly evolving landscape, confront open challenges in robustness and the integration of ML and SBI with real surveys, and look collectively toward what research in this field will — and should — look like in five years. The workshop will feature invited talks, contributed presentations, and dedicated discussion designed to foster new collaborations. It is intended for anyone working at the intersection of these themes, whether as part of a major collaboration, as a user of public data products, or as someone looking to enter this rapidly growing field.