Empirical Inference

Kernel method for percentile feature extraction

2000

Technical Report

ei


A method is proposed which computes a direction in a dataset such that a speci􏰘ed fraction of a particular class of all examples is separated from the overall mean by a maximal margin􏰤 The pro jector onto that direction can be used for class􏰣speci􏰘c feature extraction􏰤 The algorithm is carried out in a feature space associated with a support vector kernel function􏰢 hence it can be used to construct a large class of nonlinear fea􏰣 ture extractors􏰤 In the particular case where there exists only one class􏰢 the method can be thought of as a robust form of principal component analysis􏰢 where instead of variance we maximize percentile thresholds􏰤 Fi􏰣 nally􏰢 we generalize it to also include the possibility of specifying negative examples􏰤

Author(s): Schölkopf, B. and Platt, JC. and Smola, AJ.
Number (issue): MSR-TR-2000-22
Year: 2000
Day: 0

Department(s): Empirical Inference
Bibtex Type: Technical Report (techreport)

Institution: Microsoft Research

Digital: 0
Organization: Max-Planck-Gesellschaft
School: Biologische Kybernetik

Links: PDF

BibTex

@techreport{1836,
  title = {Kernel method for percentile feature extraction},
  author = {Sch{\"o}lkopf, B. and Platt, JC. and Smola, AJ.},
  number = {MSR-TR-2000-22},
  organization = {Max-Planck-Gesellschaft},
  institution = {Microsoft Research},
  school = {Biologische Kybernetik},
  year = {2000},
  doi = {}
}