Colloquium Computer Science, Sven Haase and Mandy Lange

Date:                           Monday, November 25^th 2013
Speaker:                     Sven Haase and Mandy Lange, University of Applied Sciences Mittweida (Germany), Faculty of Mathematics/Natural and Computer Sciences, Computational Intelligence Group
Room:                         5161.0267 (Bernoulliborg)
Time:                           16.00

Title: Utilization of Non-Standard Metrics in Vector Quantization

Abstract:

Vector quantization is an essential tool in data compression, clustering and classification. Prototype based vector quantization utilizes representative codebook vectors distributed in the data space for data description. Usually, the similarity or dissimilarity between data and prototypes is mathematically evaluated in terms of distance measures, e.g. the Euclidean distance as the de facto standard metric.

However, the Euclidean distance is not always the best choice. For example, if the data are grayscale images, a comparison in terms of grayscale histograms as approximations of grayscale densities might be more appropriate. Another example are gene expression data, which are usually matched by correlations, whereas l₁-norms seem to be beneficial in speech recognition when there are redundant noise features in the data.

In this talk we consider two families of non-standard metrics: The first one are general l_p-norms, which include the Euclidean distance. These norms provide a great flexibility by problem-specific adjustment of the parameter p. The talk will highlight properties and computational aspects when utilizing these distance measures. The second family of non-standard norms reflects information theoretic concepts. Specifically, divergences are under consideration as dissimilarity measures which can be seen as generalizations of correlation measures.

In the talk, both approaches are discussed and analysed in the context of unsupervised and supervised vector quantization.

Colloquium coordinators are Prof.dr. M. Aiello (e-mail : M.Aiello rug.nl ) and

Prof.dr. M. Biehl (e-mail: M.Biehl rug.nl )

http://www.rug.nl/research/jbi/news/colloquia/computerscience