Cosmological (and many other) likelihood codes are often computationally expensive to evaluate for a single parameter set and must be computed many times in the course of parameter estimation. This algorithm trains an emulator on evaluated likelihoods and uses it to estimate its values on subsequent calls. The error in the estimated likelihood can be systematically added to the posterior calculations and it speeds up parameter estimation for standard cosmological models with the Planck likelihood by a factor of between 50 and 100.
From 2004.11929: posterior distributions of the baryon density Ωbh2 with (solid blue) and without (dashed red) emulation with different samplers.
- Gihub code
- Workshop paper at ICLR 2020: Fundamental Science in the Era of AI, Aslanyan, Easther, Musoke and Price Robust posterior inference when statistically emulating forward simulations 2004.11929
- Aslanyan, Easther and Price, Learn-As-You-Go Acceleration of Cosmological Parameter Estimates, 1506.01079 or JCAP 09 (2015) 005.