Giuseppe Longo


Probing the near universe with neural networks


We outline the problems encountered in classifying large volumes of photometric data produced by a modern digital survey such as the SDSS (Sloan Digital Sky Survey), pointing out that most of these problems are produced by the degeneracy introduced in the parameter space by the lack of reliable information on the objects distance. In order to partially remove such degeneracy we implemented a supervised neural network approach to the determination of photometric redshift.

The method, even though of general validity, was fine tuned to match the characteristics of the Sloan Digital Sky Survey and takes into account the uneven distribution of measured redshifts in the SDSS spectroscopic subsample.
It consists of a two step approach. In the first step, objects are classified in nearby (z<0.25) and distant (0.25<z<0.50), and in the second step two different networks are separately trained on objects belonging to the two redshift ranges. The final results lead to a robust sigma 0.206 over the redshift range [ 0.01, 0.48].

The network was then applied to som 32 million objects from the SDSS to produce a 3-D map of the nearby universe.

The final redshifts were used as input parameter to an unsupervised neural classifier and the resulting clusters are discussed.