Efficiently learning multilayer perceptrons.

C. Bunzmann,Michael Biehl,R. Urbanczik
DOI: https://doi.org/10.1103/PHYSREVLETT.86.2166
IF: 8.6
2001-03-05
Physical Review Letters
Abstract:A learning algorithm for multilayer perceptrons is presented which is based on finding the principal components of a correlation matrix computed from the example inputs and their target outputs. For large networks our procedure needs far fewer examples to achieve good generalization than traditional on-line algorithms.
What problem does this paper attempt to address?