Nonlinear Lp approximation problems: analysis, algorithms and applications

  • Pintelon, Rik, (Administrative Promotor)

Project Details

Description

Models play an important role in a broad class of disciplines and related applications such as signal processing, system theory , economy, biomedical applications etc. The parameters of these models are determined by first collecting a large number of measurements and then solving the resulting overdetermined set of equations Ax almost equal to b in an approximate way. Determining the solution of the latter overdetermined set of equations is the kernel problem of many selection procedures.
Most model selection procedures belong to the so-called class of approximation problems. One of the best known approximation problems is the least squares problem, which belongs to the class of linear L2 problems. Other examples are the (structured) total least squares problems. Instead of minimising the sum of the squares one could think to minimise the sum of the pth powers (=Lp). For example, for p=1, this leads to the so-called least asbolute value, which is known to be more robust to outliers than the least squares approximation.
The goal of this project is to extend the solutions of the linear L2 problems to (non-)linear Lp problems.
AcronymFWOAL203
StatusFinished
Effective start/end date31/07/0131/12/05

Flemish discipline codes

  • Electrical and electronic engineering
  • Mathematical sciences

Keywords

  • system theory
  • signal processing
  • electricity
  • parameter estimation
  • signal processing