Cosmology@Home

Cosmology@Home logo

Cosmology@Home is a BOINC distributed computing project, run at the Departments of Astronomy and Physics at the University of Illinois at Urbana-Champaign.

Goals

The goal of Cosmology@Home is to compare theoretical models of the universe to the data measured to date and search for the model that best matches it.[1] Other goals may include:

Science

The goal of Cosmology@Home is to search for the model that best describes our Universe and to find the range of models that agree with the available astronomical and particle physics data. The models generated by Cosmology@home can be compared to measurements of the universe's expansion speed from the Hubble Space Telescope as well as fluctuations in the cosmic background radiation as measured by the Wilkinson Microwave Anisotropy Probe.

Method

Cosmology@Home uses an innovative way of using machine learning to effectively parallelize a large computational task that involves many inherently sequential calculations over an extremely large number of distributed computers.

For any given class of theoretically possible models of the Universe, Cosmology@Home generates tens of thousands of example Universes and packages the cosmological parameters describing these Universes as work units. Each work unit represents a single Universe. When the work unit is requested by a participating computer, this computer simulates this Universe from the Big Bang until today. At the time of writing, each such simulation takes an average of 4 hours on a current CPU. The result of this simulation is a list of observable properties of this Universe.

This result is then sent back and archived at the Cosmology@Home server. When a sufficient number of example Universes have been simulated, a machine learning algorithm called Pico,[2][3] which was developed by the project scientists of Cosmology@Home for this purpose, learns from these example calculations how to do the simulation for any Universe similar to the example Universes. The difference is that Pico takes a few milliseconds per calculation rather than several hours. Training Pico on 20,000 examples takes about 30 minutes. Once Pico is trained, it can run a full comparison of the class of models (which involves hundreds of thousands of model calculations) with the observational data in a few hours on a standard CPU.

The Cosmology@Home application is proprietary.

Milestones

See also

Notes

  1. "Letter to Cosmology@Home users". Cosmologyathome.org. Retrieved 2011-02-20.
  2. Fendt, William A.; Wandelt, Benjamin D. (2007). "PICO: Parameters for the impatient cosmologist". The Astrophysical Journal (The American Astronomical Society) 654: 2–11. arXiv:astro-ph/0606709. Bibcode:2007ApJ...654....2F. doi:10.1086/508342
  3. Fendt, William A.; Wandelt, Benjamin D. "Computing High accuracy power spectra with Pico". arXiv:0712.0194
  4. "Beta testing!". Cosmologyathome.org. Retrieved 2011-02-20.

External links

This article is issued from Wikipedia - version of the Saturday, May 16, 2015. The text is available under the Creative Commons Attribution/Share Alike but additional terms may apply for the media files.