Dresden 2003 – scientific programme

Parts | Days | Selection | Search | Downloads | Help

DY: Dynamik und Statistische Physik

DY 46: Poster

DY 46.5: Poster

Thursday, March 27, 2003, 15:30–18:00, P1

Approximating the stationary probability density from on-line presented data — •Michael Schindler — Universität Augsburg

For many scientific and technical purposes it is necessary to analyse clustered data in many dimensions. There are some algorithms available that can find an approximation for the stationary probability density of such data. Their main drawback is, however, that they need randomised data as input. Whenever dealing with “on-line” (that is sequentially presented temporal correlated) data, such as data having a stochastic process as generator, they may fail to find a representation of the stationary properties.

Starting with an algorithm that can be formulated in a probabilistic sound way as a Maximum-Likelihood approximator we show in detail the problems that arise during on-line learning and present an approach to overcome them.