Neighbor Embedding by Soft Kendall Correlation

Strickert, Marc; Hüllermeier, Eyke

Neighbor Embedding by Soft Kendall Correlation

Files

073-077.pdf (441.11 KB)

Date

2013

Authors

Strickert, Marc
Hüllermeier, Eyke

Publisher

The Eurographics Association

Abstract

Correlation-based embedding of complex data relationships in a Euclidean space is studied. The proposed soft formulation of Kendall correlation allows for gradient-based optimization of scatter point neighborhood relationships for reconstructing original data neighbors. The approach is able to handle asymmetric data relations provided in the form of a general scoring matrix. Scale and shift invariance properties of correlation help circumventing typical embedding distortion artefacts in dimension reduction and data embedding scenarios.

        @inproceedings{:10.2312/PE.EuroVisShort.EuroVisShort2013.073-077
,
booktitle = {EuroVis - Short Papers
},
editor = {Mario Hlawitschka and Tino Weinkauf
},
title = {{Neighbor Embedding by Soft Kendall Correlation
}},
author = {Strickert, Marc and 
Hüllermeier, Eyke
},
year = {2013
},
publisher = {The Eurographics Association
},
ISBN = {978-3-905673-99-9
},
DOI = {/10.2312/PE.EuroVisShort.EuroVisShort2013.073-077
}
}