I have another website that more generally describes my research area, talks I have given, publications and non-academic interests. It is located here other website
This webpage describes the process of creating a neural network to distinguish between quasars and point-like objects using photometric properties of the data.
Context
To measure properties of our universe, galaxy surveys use large telescopes to observe the light of millions of objects. By measuring the distribution of different classes of objects we can infer properties of the universe such as it's expansion rate.
Information about an object is derived from the properties of emitted light we observe. Properties are measured via two different methods, photometric and spectroscopic analysis.
-
Photometic observations measure the amount of light passing through coloured filters. This measures the intensity of light but only as a function of a few wavelengths. This method permits multiple observations at one time making it relatively inexpensive to gather broad information on a large number of objects.
-
In spectroscopic observations the light travels through a spectrometer and the continuous intensity of light as a function of wavelength is recovered. For each object more detailed information is measured but it is a more expensive procedure as fewer objects can be observed at one time.
Spectroscopic observations are more expensive but more informative, therefore photometric observations are carried out prior to spectroscopic observations inorder to get an idea of the location of interesting objects that warrant targeting with our spectroscopic survey.
Aim
Given a list of objects from a photometric survey with similar photometric attributes that we cannot use to classify the objects by making simple cuts in the data, train a neural network to identify quasars from other objects.