Journal Article

<span class="smallCaps">cosmonet</span>: fast cosmological parameter estimation in non-flat models using neural networks

T. Auld, M. Bridges and M. P. Hobson

in Monthly Notices of the Royal Astronomical Society

Published on behalf of The Royal Astronomical Society

Volume 387, issue 4, pages 1575-1582
Published in print July 2008 | ISSN: 0035-8711
Published online July 2008 | e-ISSN: 1365-2966 | DOI:
cosmonet: fast cosmological parameter estimation in non-flat models using neural networks

More Like This

Show all results sharing this subject:

  • Astronomy and Astrophysics


Show Summary Details


We present a further development of a method for accelerating the calculation of cosmic microwave background (CMB) power spectra, matter power spectra and likelihood functions for use in cosmological Bayesian inference. The algorithm, called cosmonet, is based on training a multilayer perceptron neural network. We demonstrate the capabilities of cosmonet by computing CMB power spectra (up to ℓ= 2000) and matter transfer functions over a hypercube in parameter space encompassing the 4 σ confidence region of a selection of CMB [Wilkinson Microwave Anisotropy Probe (WMAP) + high-resolution experiments] and large-scale structure surveys [2dF and Sloan Digital Sky Survey (SDSS)]. We work in the framework of a generic seven parameter non-flat cosmology. Additionally, we use cosmonet to compute the WMAP 3 yr, 2dF and SDSS likelihoods over the same region. We find that the average error in the power spectra is typically well below cosmic variance for spectra, and experimental likelihoods calculated to within a fraction of a log unit. We demonstrate that marginalized posteriors generated with cosmonet spectra agree to within a few per cent of those generated by camb parallelized over four CPUs, but are obtained two to three times faster on just a single processor. Furthermore, posteriors generated directly via cosmonet likelihoods can be obtained in less than 30 min on a single processor, corresponding to a speed up of a factor of ∼32. We also demonstrate the capabilities of cosmonet by extending the CMB power spectra and matter transfer function training to a more generic 10 parameter cosmological model, including tensor modes, a varying equation of state of dark energy and massive neutrinos. Finally, we demonstrate that using cosmonet likelihoods directly, the sampling strategy adopted by cosmomc is highly suboptimal. We find the generic bayesys sampler to be a further ∼10 times faster, yielding 20 000 post burn-in samples in our seven parameter model in just 3 min on a single CPU. cosmonet and interfaces to both cosmomc and bayesys are publically available at

Keywords: methods: data analysis; methods: statistical; cosmic microwave background

Journal Article.  5435 words.  Illustrated.

Subjects: Astronomy and Astrophysics

Full text: subscription required

How to subscribe Recommend to my Librarian

Users without a subscription are not able to see the full content. Please, subscribe or login to access all content.