Λ -conditional flow matching: scale-aware representation learning for cosmology with flow matching

Abstract In modern cosmology, high-resolution simulations of the cosmic web generate petabytes of data, creating a critical need for dimensionality reduction schemes that preserve scientific information for downstream analysis. In this work, we demonstrate that flow matching -based generative models can learn compact, semantically rich latent representations of field level cold dark matter simulation data without supervision. Our model, Λ -Conditional Flow Matching, learns representations 32 × smaller than the raw field data, usable for field level reconstruction and synthetic data generation. We show that an additional 64 × compression is possible, yielding summary statistics can be used to estimate the cosmological parameters, Ω m and σ 8 to within 5% accuracy. Finally, we show that through the use of a time-dependent masking scheme, our model also learns interpretable, scale-aware representations, in which different channels of the compressed representation correspond to features at different cosmological scales.