Introduction:

CosmoNet allows for fast and efficient estimation of CMB power spectra (TT, TE, EE and BB) and WMAP likelihood values. This allows sampling programs such as cosmomc to perform parameter estimation in minutes on a laptop computer as opposed to the hours of supercomputer time normally needed. We have trained perceptron multilayer neural networks for flat and non-flat cosmological models in the 6(7) dimensional parameter space: omega_bh^2, omega_ch^2, (omega_k), theta, tau, n_s and ln A_s and plan to release networks for other models in due course. The neural networks are used to interpolate the CMB power spectra and likelihoods for all models within the region covered by our training region. If the sampler attempts to find a model outside this region it will simply call CAMB and the WMAP likelihood code in the normal fashion.

CosmoNets neural network maps are very simple and typically contain a small number of thousands of parameters (weights). CosmoNet weights are available on this page enabling anyone to write their own implementation of the neural networks, in a language of their own choosing. See the 'build your own' section below.

CosmoNet and CosmoMC:

We have implemented cosmonet into the latest version of cosmomc, the archive files for which are available: 6 param and 7 param.

Build Your Own CosmoNet:

A receipe for building your own cosmological neural networks is given here.

Network weights for the flat cosmological models in the 6 dimensional parameter space are found here.

Weights for other, non-flat models will shortly be released on this site.

Web page maintained by Michael Bridges.