ML Anthology

Robust Formulations for Training Multilayer Perceptrons

Tommi Kärkkäinen, Erkki Heikkola
NeCo 2004 pp. 837-862
doi:10.1162/089976604322860721 /neco/2004/karkkainen2004neco-robust/

Abstract

The connection between robust statistical estimates and nonsmooth optimization is established. Based on the resulting family of optimization problems, robust learning problem formulations with regularization-based control on the model complexity of the multilayer perceptron network are described and analyzed. Numerical experiments for simulated regression problems are conducted, and new strategies for determining the regularization coefficient are proposed and evaluated.

NeCo Semantic Scholar

Cite

Text

Kärkkäinen and Heikkola. "Robust Formulations for Training Multilayer Perceptrons." Neural Computation, 2004. doi:10.1162/089976604322860721

Markdown

[Kärkkäinen and Heikkola. "Robust Formulations for Training Multilayer Perceptrons." Neural Computation, 2004.](https://mlanthology.org/neco/2004/karkkainen2004neco-robust/) doi:10.1162/089976604322860721

BibTeX

@article{karkkainen2004neco-robust,
  title     = {{Robust Formulations for Training Multilayer Perceptrons}},
  author    = {Kärkkäinen, Tommi and Heikkola, Erkki},
  journal   = {Neural Computation},
  year      = {2004},
  pages     = {837-862},
  doi       = {10.1162/089976604322860721},
  volume    = {16},
  url       = {https://mlanthology.org/neco/2004/karkkainen2004neco-robust/}
}