Neural networks and separation of background and foregrounds in astrophysical sky maps
Carlo Baccigalupi, Luigi Bedini, C. Burigana, G. de Zotti, A. Farusi, Davide Maino, MICHELE MARIS, Francesca Perrotta, Emanuele Salerno, Luigi Toffolatti, Anna Tonazzini · arXiv (Cornell University) · 2000
The Independent Component Analysis (ICA) algorithm is implemented as a neuralnetwork for separating signals of different origin in astrophysical sky maps.Due to its self-organizing capability, it works without prior assumptions onthe signals, neither on their frequency scaling, nor on the signal mapsthemselves; instead, it learns directly from the input data how to separate thephysical components, making use of their statistical independence. To test thecapabilities of this approach, we apply the ICA algorithm on sky patches, takenfrom simulations and observations, at the microwave frequencies, that are goingto be deeply explored in a few years on the whole sky, by the MicrowaveAnisotropy Probe (MAP) and by the {\\sc Planck} Surveyor Satellite. The maps areat the frequencies of the Low Frequency Instrument (LFI) aboard the {\\scPlanck} satellite (30, 44, 70 and 100 GHz), and contain simulated astrophysicalradio sources, Cosmic Microwave Background (CMB) radiation, and Galacticdiffuse emissions from thermal dust and synchrotron. We show that the ICAalgorithm is able to recover each signal, with precision going from 10 0.000000or theGalactic components to percent for CMB; radio sources are almost completelyrecovered down to a flux limit corresponding to $0.7\\sigma_{CMB}$, where$\\sigma_{CMB}$ is the rms level of CMB fluctuations. The signal recoveringpossesses equal quality on all the scales larger then the pixel size. Inaddition, we show that the frequency scalings of the input signals can bepartially inferred from the ICA outputs, at the percent precision for thedominant components, radio sources and CMB.