Reproduce “Recovering Missing Degrees of Freedom in Simulation-Trained Field-Level Generative Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Masato Shirasaki, Shiro Ikeda
The authors add physics-motivated latent controls to a continuous normalizing flow for weak-lensing convergence maps, addressing realization-to-realization variance shrinkage without additional training simulations. The controls are optimized after training, with latent variables sampled during inference. On two tested distribution shifts, the dominant latent moves inferred power-spectrum responses toward simulations and lowers mismatch for most realizations, but covariance recovery remains partial.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.