Repository navigation
I want to know waht's going on #18
Description
Activity
Claude finished @EiffL's task in 41s —— View job
What SHINE Does
- Gather context about the repository
- Provide explanation of what SHINE does
SHINE (SHear INference Environment) is a JAX-powered framework for probabilistic weak gravitational lensing shear estimation — a core measurement in cosmology used to study dark matter and dark energy.
The Core Problem
When light from distant galaxies passes near massive objects (galaxy clusters, dark matter halos), it gets deflected, causing the galaxy images to appear slightly distorted. Measuring this gravitational shear (the subtle stretching/rotation of galaxy shapes) lets astronomers map the distribution of matter in the universe.
Traditional approaches measure galaxy ellipticities and apply statistical corrections for instrumental effects. SHINE takes a fundamentally different, more principled approach.
SHINE's Approach: Bayesian Forward Modeling
Instead of measuring and correcting, SHINE:
- Builds a generative forward model of the sky: renders galaxy images (Sersic profiles) via JAX-GalSim, convolves them with the telescope PSF, adds noise
- Compares the forward model to observed pixel data to evaluate a likelihood
- Runs Bayesian inference (MAP optimization or full MCMC/NUTS via NumPyro) to recover the posterior distribution of shear parameters
This treats shear measurement as a proper Bayesian inverse problem, avoiding the multiplicative/additive bias corrections that plague traditional methods.
Technology Stack
Component Role JAX JIT compilation, autodiff, GPU support NumPyro Probabilistic programming, NUTS/HMC sampling JAX-GalSim Differentiable galaxy rendering & PSF convolution Pydantic YAML config validation Current Status
The first instrument backend is
shine.euclid— a full pipeline for Euclid VIS multi-exposure data:- Reads quadrant FITS files, PSF grids, and MER source catalogs
- Filters sources by SNR, quality flags, and size
- Renders scenes with spatially-varying PSFs across multiple exposures
- Infers shared shear
(g1, g2)+ per-source morphology parameters
The project is in early/alpha development from the CosmoStat Lab (CEA/CNRS), with plans to extend to LSST, MeerKAT, and non-parametric galaxy morphology models.
