Group items tagged

Modern MCMC method for cosmological parameter estimation. "While Metropolis-Hastings is constrained by overheads, CosmoHammer is able to accelerate the sampling process from a wall time of 30 hours on a single machine to 16 minutes by the efficient use of 2048 cores. Such short wall times for complex data sets opens possibilities for extensive model testing and control of systematics."

A sampler, new in both algorithm and code, from Hogg and co. They claim speed ups over Metropolis Hastings and easy shared-memory parallelism using an ensemble method.

Hints of something interesting or data artefact? Also: http://arxiv.org/pdf/1207.4781v1.pdf

I've not read the whole thing (it's 44 pages!) but Yu Lu is, IMO, doing the Right Thing in this field - he takes a Semi Analytic Model of galaxy formation, and *actually fits* all the parameters to the data (in this case, the observed K-band luminosity function). Some parameters are well-constrained (implying we may have learnt something about galaxies), while others show strong degeneracies, indicating what new physics needs to be included. Seems like the models in most of the parameter volume fail to predict some other datasets, giving more clues on how to improve the model. The key thing is that by doing the inference properly, Lu has elevated SAM study to a quantified learning process.

CosmoMC gets upgraded by the Cavendish inference team (Hobson, Feroz, and now Graff) - they approximate complex likelihood functions with neural networks, which are then much much faster to evaluate. Could be a real time-saver