Cosmology beyond the lamp post

Cosmologists strive to uncover the mysteries of the origin, composition, evolution, and fate of the cosmos from all the information the sky has to offer. I will discuss how machine learning methods have made previously intractable problems tractable. The resulting capabilities are revolutionizing our ability to confront computational models with data, laying the ground for new discoveries. Recent breakthroughs include AI-accelerated simulations that generate galaxy catalogs with remarkable fidelity, techniques to evaluate the information content of data within non-Gaussian and non-linear settings, generative models that reconstruct the universe's initial conditions from today's complex cosmic web, and methods that enable rapid Bayesian model comparison. Recent work addresses more general challenges in scientific inference: automatically identifying informative summaries from data, estimating model dimensionality, and making inferences robust against model misspecification. Together, these methods enable end-to-end forward modeling and inference pipelines connecting fundamental physics with observations. At the same time cosmology provides an ideal testbed for method development in machine learning: vast open datasets spanning diverse physical regimes, well-defined theoretical predictions for validation, and problems that stress-test inference methods at scale. This symbiotic relationship advances both fundamental methodology and our understanding of the universe's detailed history from the Big Bang to the present day