Dresden 2003 – scientific programme

Parts | Days | Selection | Search | Downloads | Help

DY: Dynamik und Statistische Physik

DY 36: Nonlinear dynamics II

DY 36.4: Talk

Wednesday, March 26, 2003, 17:15–17:30, G\"OR/226

Structure Detection Using Complexity Measures and Neural Networks — •René Pompl — Centre for Interdisciplinary Plasma Science / Max-Planck-Institut für extraterrestrische Physik, Gießenbachstraße, 85741 Garching

The detection of structures in the presence of noise is a recurring problem in data analysis. Structural complexity measures like the scaling index method are a useful tool due to their ability to quantitatively characterize the local scaling properties in point distributions of arbitrary dimension. Neural networks are also widely used, since they are capable to adapt to a problem in a self-organising manner. This contribution investigates whether a combination of both approaches leads to a better structure detection.

Thereto a network architecture, which models the scaling index method, together with a supervised learning algorithm is presented. Due to the general formulation such a network is applicable to time series and high dimensional point distributions. The properties of such a network are investigated using three-dimensional point distributions, namely artificially generated grey-value images. As a result a clear improvement of the detection rate could be observed.