direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

Es gibt keine deutsche Übersetzung dieser Webseite.

Event Calendar

Montag, 06.06.2016

BIMoS Day "Probabilistic and Bayesian Data Modeling"

Art der Veranstaltung:
Vortragsreihe, -programm, Veranstaltungsreihe

BIMoS Day "Probabilistic and Bayesian Data Modeling"

Speaker: Prof. Dr. Manfred Opper (TUB)

When: Monday, June 6th, 4 pm - 6 pm (16.00 Uhr s.t.)

Where: H 3005, TUB Main Building


Probabilistic data models play an important role in many areas of research such as statistics, machine learning, artificial intelligence and signal processing. These models assume that data is generated at random from certain, often highly complex probability distributions. They encode both the regularities and the noise in the data. Given a set of observed data, one tries to €˜learn the unknown distribution in order to be able make good predictions. In a Bayesian approach to statistics this is achieved by assuming that all unknown quantities, such as the parameters of the model, should be treated as random variables. Hence, one needs to specify also probability distributions over model parameters. These distributions should encode the modeler's prior knowledge. Predictions are based on the posterior distribution which gives the probability of parameters conditioned on the observed data. It combines prior knowledge and the information given by observations. This approach does not only provide likely explanations of the data but also gives measures of uncertainty on our predictions. In addition, one gets a conceptually simple method for comparing different models. In this talk, I will give an introduction to Bayesian models starting with the simple problem of linear regression and ending with nonparametric models where the complexity of the model is allowed to grow with the number of observations.

  • office@bimos.tu-berlin.de,
H 3005, Straße des 17. Juni 135, 10623 Berlin
16:00 - 18:00

Event Calendar

Event Calendar

«Oktober 22»

Zusatzinformationen / Extras


Schnellnavigation zur Seite über Nummerneingabe