Regensburg 2004 – scientific programme

Parts | Days | Selection | Search | Downloads | Help

DY: Dynamik und Statistische Physik

DY 46: Poster

DY 46.12: Poster

Thursday, March 11, 2004, 16:00–18:00, Poster D

The Dynamics of Learning Vector Quantization — •Michael Biehl^1,2, Ansgar Freking², and Georg Reents² — ¹Institute for Mathematics and Computing Science, University of Groningen, P.O. Box 800, NL-9700 AV Groningen — ²Institut fuer Theoretische Physik und Astrophysik, Universitaet Wuerzburg, Am Hubland, D-97074 Wuerzburg

Learning Vector Quantization (LVQ) is a widely used method for the implementation of classification schemes, in particular in the context of structured (No-dqclusteredNo-dq) data. Here, we analyse LVQ by means of the statistical physics theory of on-line learning in high-dimensional weight space.

We show how variations of LVQ can be interpreted as interpolating between No-dqunsupervised clusteringNo-dq (without a teacher) and the supervised learning of a rule. The focus is on the phenomen of symmetry breaking phenomena and the role of initial conditions for the learning process.