Adaptive relevance matrices in LVQ

Link zum Paper

• Funfact: Dieses Paper ist von Hammer.

Summary

• Developed an extension of LVQ to an adaptive matrix of relevances which parameterizes a general Euclidean metric and accounts for pairwise correlations of features

• Apprach allows to choose a metric, the measures of the initial papers could be brought in.

• To define a valid metric, symmetry and positive definiteness of a matrix has to be enforced.

• Ähnlicher Ansatz wie in Geometrie Verarbeitung mit der ersten Fundamentalform für Metrik

• Experiment with artifical data and comparisons to other LVQ results.

Tasks and Ideas

• Train GMLVQ with metric of the initial paper and compare results.
• Transform fairness measures to a metric that statisfies conditions of the learning matrix.