Back to Volume
Paper: Using Unlabeled Data to Improve the Automated Prediction of Stellar Atmospheric Parameters
Volume: 281, Astronomical Data Analysis Software and Systems XI
Page: 405
Authors: Solorio, Thamar; Fuentes, Olac
Abstract: With the recent availability of large sky surveys has come a need to automatically analyze these data. Several algorithms based on machine learning have been proposed for this task, providing results that are comparable to those obtained by human experts. However, these algorithms normally require a large set of manually labeled data or training; if the available training set is small, performance is usually poor. In this work we present an approach to automated astronomical classification that performs well in cases where the training set is small by taking advantage of the information contained in the test set, which is usually large. We have applied our method to the problem of predicting stellar atmospheric parameters from spectral information. Experimental results show that our method provides an error reduction of up to 25%.
Back to Volume