Neural computation as a tool for galaxy classification: methods and examples

We apply and compare various Artificial Neural Network (ANN) and otheralgorithms for automatic morphological classification of galaxies. The ANNs arepresented here mathematically, as non-linear extensions of conventionalstatistical methods in Astronomy. The methods are illustrated using differentsubsets Artificial Neural Network (ANN) and other algorithms for automaticmorphological classification of galaxies. The ANNs are presented heremathematically, as non-linear extensions of conventional statistical methods inAstronomy. The methods are illustrated using different subsets from the ESO-LVcatalogue, for which both machine parameters and human classification areavailable. The main methods we explore are: (i) Principal Component Analysis(PCA) which tells how independent and informative the input parameters are.(ii) Encoder Neural Network which allows us to find both linear (PCA-like) andnon-linear combinations of the input, illustrating an example of unsupervisedANN. (iii) Supervised ANN (using the Backpropagation or Quasi-Newtonalgorithms) based on a training set for which the human classification isknown. Here the output for previously unclassified galaxies can be interpretedas either a continuous (analog) output (e.g. $T$-type) or a Bayesian {\it aposteriori} probability for each class. Although the ESO-LV parameters aresub-optimal, the success of the ANN in reproducing the human classification is2 $T$-type units, similar to the degree of agreement between two human expertswho classify the same galaxy images on plate material. We also examine theaspects of ANN configurations, reproducibility, scaling of input parameters andredshift information.