cosmonet : fast cosmological parameter estimation in non‐flat models using neural networks

We present a further development of a method for accelerating the calculationof CMB power spectra, matter power spectra and likelihood functions for use incosmological Bayesian inference. The algorithm, called {\sc CosmoNet}, is basedon training a multilayer perceptron neural network. We compute CMB powerspectra (up to $\ell=2000$) and matter transfer functions over a hypercube inparameter space encompassing the $4\sigma$ confidence region of a selection ofCMB (WMAP + high resolution experiments) and large scale structure surveys (2dFand SDSS). We work in the framework of a generic 7 parameter non-flatcosmology. Additionally we use {\sc CosmoNet} to compute the WMAP 3-year, 2dFand SDSS likelihoods over the same region. We find that the average error inthe power spectra is typically well below cosmic variance for spectra, andexperimental likelihoods calculated to within a fraction of a log unit. Wedemonstrate that marginalised posteriors generated with {\sc CosmoNet} spectraagree to within a few percent of those generated by {\sc CAMB} parallelisedover 4 CPUs, but are obtained 2-3 times faster on just a \emph{single}processor. Furthermore posteriors generated directly via {\sc CosmoNet}likelihoods can be obtained in less than 30 minutes on a single processor,corresponding to a speed up of a factor of $\sim 32$. We also demonstrate thecapabilities of {\sc CosmoNet} by extending the CMB power spectra and mattertransfer function training to a more generic 10 parameter cosmological model,including tensor modes, a varying equation of state of dark energy and massiveneutrinos. {\sc CosmoNet} and interfaces to both {\sc CosmoMC} and {\scBayesys} are publically available at {\ttwww.mrao.cam.ac.uk/software/cosmonet}.Comment: 8 pages, submitted to MNRA