Mittwoch, 19. Mrz 2014, 15:00 - 16:00 iCal

CS-Colloquium: Prof. Dr. Katharina Morik

Big Data Analytics in Astrophysics

Fakultät für Informatik - HS3
Währinger Straße 29, 1090 Wien


Big Data has become an increasingly popular term since 2011. Its facets, volume, variety, and velocity, have led to new storage and computation methods and hardware architectures. These developments may support but cannot replace the analysis of the data. Analytics is necessary in order to reap the data’s benefit. In this talk, analysis methods for large volume data and for real-time velocity data are presented and illustrated by astrophysical experiments.

Astrophysical data can be so large, that they are faster to be shipped than to be broadcasted via satellite. For instance, the IceCube experiment at the South Pole produces 1 Terabyte of data per day. A satellite would need about 10 years to communicate the data of a year, but, stored on hard disk, they are shipped within 28 days from the South Pole to the University of Wisconsin. IceCube searches for neutrinos within a strongly biased distribution of signal and background events: 93,000 neutrinos versus 17.48 million background events. The challenge for Big Data Analytics is to detect the true neutrino events in the data and to carefully evaluate the learned model. Together with Dortmund colleagues from the IceCube collaboration, Wolfgang Rhode and Tim Ruhe, we could significantly increase the rate of detected true neutrinos and recognize a neutrino event at 1 Peta eV. The key to success was the preprocessing which automatically determined the right feature space.

Another physical experiment measures air showers by two Cherenkov telescopes, MAGIC and FACT at La Palma. It aims at detecting gamma-rays in order to better understand active galaxies. From gamma-rays, their source can be calculated, but they come along with several millions observations of other particles whose paths are influenced by intergalactic magnetic fields so that they do not allow to infer their source. Again, the challenge for Big Data Analytics is to find the needle in the hay stack and to justify the finding. An approach to process data streams in real-time and to combine this with the map-reduce approach is illustrated by inspecting the FACT data.

Zur Webseite der Veranstaltung


Fakultät für Informatik


Werner Schröttner
Fakultät für Informatik