Knowledge Discovery in Mega-Spectra Archives

The recent progress of astronomical instrumentation resulted in the constructionof multi-object spectrographs with hundreds to thousands of micro-slits or opticalfibers allowing the acquisition of tens of thousands of spectra of celestial objectsper observing night. Currently there are several spectroscopic surveys containingmillions of spectra and much larger are in preparation. Most of the large-scalesurveys are processed spectrum by spectrum in order to estimate physical param-eters of individual objects. The parameters obtained are then used to constructthe better models of space-kinematic structure and evolution of the Universe orits subsystems. Such surveys are, however, very good source of homogenized, pre-processed data for application of machine learning techniques and advanced statis-tical processing common in Astroinformatics. We present challenges of knowledgediscovery process applied to large spectroscopic surveys as well as memory spaceand processing speed demands of current machine learning methods, requiring BigData techniques.