31 August 2026 to 4 September 2026
US/Pacific timezone
All in-person registration fee waivers have now been claimed.

Generative Cosmological Emulators as Fluid Flows in Data Space

31 Aug 2026, 17:30
1h 30m
QI Courtyard

QI Courtyard

Speaker

Sandip Roy (University of California San Diego)

Description

Cosmological simulations are too computationally expensive to support exhaustive scans over dark matter parameter space. Generative models, including diffusion models and flow models, offer a promising route to simulation emulation using existing suites of numerical simulations. In this talk, I will describe diffusion and flow-based generative models as transport processes: they evolve probability densities through an abstract, high-dimensional data space according to equations closely related to continuity and flow equations derived from the boltzmann equation. This viewpoint provides a natural bridge to cosmological simulation, where density fields and particle distributions are likewise evolved under physically constrained dynamics. I will show how this connection can be used to inject theoretical priors into generative emulators. I will demonstrate the approach on both density-field emulation and point-cloud representations of cosmological boxes, with the goal of building faster and more physically informed surrogates for exploring dark matter parameter space.

Author

Sandip Roy (University of California San Diego)

Presentation materials

There are no materials yet.