Download PDFOpen PDF in browser

A Sparse Representation of High-Dimensional Input Spaces Based on an Augmented Growing Neural Gas

11 pagesPublished: September 29, 2016

Abstract

\noindent The growing neural gas (GNG) algorithm is an unsupervised learning method that is able to approximate the structure of its input space with a network of prototypes. Each prototype represents a local input space region and neighboring prototypes in the GNG network correspond to neighboring regions in input space. Here we address two problems that can arise when using the GNG algorithm. First, the GNG network structure becomes less and less meaningful with increasing dimensionality of the input space as typical distance measures like the Euclidean distance loose their expressiveness in higher dimensions. Second, the GNG itself does not provide a form of output that retains the discovered neighborhood relations when compared with common distance measures. We show that a GNG augmented with {\em local input space histograms} can mitigate both of these problems. We define a sparse vector representation as output of the augmented GNG that preserves important neighborhood relations while pruning erroneous relations that were introduced due to effects of high dimensionality.

Keyphrases: curse of dimensionality, Growing Neural Gas, local input space histograms, sparse representation

In: Christoph Benzmüller, Geoff Sutcliffe and Raul Rojas (editors). GCAI 2016. 2nd Global Conference on Artificial Intelligence, vol 41, pages 303--313

Links:
BibTeX entry
@inproceedings{GCAI2016:Sparse_Representation_of_High_Dimensional,
  author    = {Jochen Kerdels and Gabriele Peters},
  title     = {A Sparse Representation of High-Dimensional Input Spaces Based on an  Augmented Growing Neural Gas},
  booktitle = {GCAI 2016. 2nd Global Conference on Artificial Intelligence},
  editor    = {Christoph Benzm\textbackslash{}"uller and Geoff Sutcliffe and Raul Rojas},
  series    = {EPiC Series in Computing},
  volume    = {41},
  pages     = {303--313},
  year      = {2016},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2398-7340},
  url       = {https://easychair.org/publications/paper/s2},
  doi       = {10.29007/jgjt}}
Download PDFOpen PDF in browser