Outlier Analysis in BP/RP Spectral Bands

摘要

Most astronomic databases include a certain amount of exceptional values that are generally called outliers. Isolating and analysing these "outlying objects" is important to improve the quality of the original dataset, to reduce the impact of anomalous observations, and most importantly, to discover new types of objects that were hitherto unknown because of their low frequency or short lifespan. We propose an unsuper-vised technique, based on artificial neural networks and combined with a specific study of the trained network, to treat the problem of outliers management. This work is an integrating part of the GAIA mission of the European Space Agency.