Retraining of the SAMANN Network

1
Retraining of the SAMANN Network
Viktor Medvedev, Gintautas Dzemyda Viktor.m,
Dzemyda_at_ktl.mii.lt Institute of Mathematics and
Informatics Vilnius, Lithuania
32nd International Conference on Current Trends
in Theory and Practice of Computer
Science Student Research Forum January 21 - 27,
2006 Merin, Czech Republic
SOFSEM 2006, SRF
2
Key words visualization, multidimensional data,
Sammons mapping, SAMANN neural network,
retraining of the network

• Multidimensional data. Observations from
  real-world problems are often highdimensional
  vectors. The problem is to discover knowledge in
  the set of multidimensional points.
• Visualization is a powerful tool in data
  analysis. It makes easier the understandability
  and perception of data.
• Sammons mapping, multidimensional scaling,
  principal components
• Sammons mapping a well-known procedure for
  mapping data from a high-dimensional space onto a
  lower-dimensional one.
• A neural network for sammons projection

SOFSEM 2006, SRF
3

• SAMANN a specific backpropagation algorithm to
  train a multilayer feed-forward artificial neural
  network (SAMANN) to perform the Sammons
  nonlinear projection in an unsupervised way.
• The network is able to project new patterns after
  training.
• Retraining of the network. While working with
  large data amounts there may appear a lot of new
  vectors.
• Strategies for retraining the network. Some
  strategies for retraining the network that
  realizes multidimensional data visualization have
  been proposed.
• One of the proposed strategies enables us to
  attain good visualization results in a very short
  time as well as to get smaller visualization
  errors and to improve the accuracy of projection
  as compared to other strategies.

SOFSEM 2006, SRF
4
Thank you for your attention
SOFSEM 2006, SRF