Recycling on a Cosmic Scale: Extracting New Information from Old Data Sets
Publikation: Bog/antologi/afhandling/rapport › Ph.d.-afhandling › Forskning
Standard
Recycling on a Cosmic Scale : Extracting New Information from Old Data Sets. / Stensbo-Smidt, Kristoffer.
Department of Computer Science, Faculty of Science, University of Copenhagen, 2016.Publikation: Bog/antologi/afhandling/rapport › Ph.d.-afhandling › Forskning
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - BOOK
T1 - Recycling on a Cosmic Scale
T2 - Extracting New Information from Old Data Sets
AU - Stensbo-Smidt, Kristoffer
PY - 2016
Y1 - 2016
N2 - Astronomy and astrophysics are entering a data-rich era. Large surveys have, quite literally,seen the light in the past decade, with more and larger telescopes to follow in thecoming years. Data is now so abundant that making use of all the information is a difficulttasks. This thesis sets out from the assumption that there is more to gain from availabledata sets – new information from old data. Three contributions in this direction are considered.Firstly, a novel texture descriptor for parametrising galaxy morphology is presented. Ituses the shape index and curvedness of local regions in images of galaxies and condensesinformation about the local structure to a single value. It is argued that this value can beinterpreted as indicating regions of morphological interest, for example regions of newlyformed stars, of gas and dust, spiral arms etc. The descriptor is shown to extract informationabout a galaxy’s specific star formation rate from its images that the usual spectraenergy distribution (SED) fitting misses.Secondly, a method to evaluate the information content of various features for a giventask is introduced. Selecting the right features, for example colours or magnitudes, for aspecific task can be difficult and often relies on which have been used traditionally. Withcurrent and future surveys giving researchers access to hundreds of features, it is timeto challenge old assumptions on which to use. A completely general method for featureselection is introduced and shown to increase accuracy of both redshift and specific starformation estimations.Thirdly, the problem of quality assessment of quasar candidates is considered. Detectionpipelines searching the sky for quasars produce thousands of candidates, many ofwhich can be discarded with simple checks. The rest, however, cannot, and images of thesecandidates must be manually inspected and evaluated. Still, more than 90% of these canbe false positives, wasting precious time for researchers and forcing a limitation of thescopes of the detection pipelines. A set of features based on image analysis is presentedand shown to be able to detect the most common situations of false positive quasar candidates.Incorporation of the derived features into a machine learning frameworks is reviewedand future directions are discussed.
AB - Astronomy and astrophysics are entering a data-rich era. Large surveys have, quite literally,seen the light in the past decade, with more and larger telescopes to follow in thecoming years. Data is now so abundant that making use of all the information is a difficulttasks. This thesis sets out from the assumption that there is more to gain from availabledata sets – new information from old data. Three contributions in this direction are considered.Firstly, a novel texture descriptor for parametrising galaxy morphology is presented. Ituses the shape index and curvedness of local regions in images of galaxies and condensesinformation about the local structure to a single value. It is argued that this value can beinterpreted as indicating regions of morphological interest, for example regions of newlyformed stars, of gas and dust, spiral arms etc. The descriptor is shown to extract informationabout a galaxy’s specific star formation rate from its images that the usual spectraenergy distribution (SED) fitting misses.Secondly, a method to evaluate the information content of various features for a giventask is introduced. Selecting the right features, for example colours or magnitudes, for aspecific task can be difficult and often relies on which have been used traditionally. Withcurrent and future surveys giving researchers access to hundreds of features, it is timeto challenge old assumptions on which to use. A completely general method for featureselection is introduced and shown to increase accuracy of both redshift and specific starformation estimations.Thirdly, the problem of quality assessment of quasar candidates is considered. Detectionpipelines searching the sky for quasars produce thousands of candidates, many ofwhich can be discarded with simple checks. The rest, however, cannot, and images of thesecandidates must be manually inspected and evaluated. Still, more than 90% of these canbe false positives, wasting precious time for researchers and forcing a limitation of thescopes of the detection pipelines. A set of features based on image analysis is presentedand shown to be able to detect the most common situations of false positive quasar candidates.Incorporation of the derived features into a machine learning frameworks is reviewedand future directions are discussed.
UR - https://soeg.kb.dk/permalink/45KBDK_KGL/fbp0ps/alma99122536952205763
M3 - Ph.D. thesis
BT - Recycling on a Cosmic Scale
PB - Department of Computer Science, Faculty of Science, University of Copenhagen
ER -
ID: 172265953