Knowledge Driven Dimension Reduction For Clustering
As A.I. algorithms are applied to more complex domains that involve high dimensional data sets there is a need to more saliently represent the data. However, most dimension reduction approaches are driven by objective functions that may not or only partially suit the end users requirements. In this work, we show how to incorporate general-purpose domain expertise encoded as a graph into dimension reduction in way that lends itself to an elegant generalized eigenvalue problem. We call our approach Graph-Driven Constrained Dimension Reduction via Linear Projection (GCDR-LP) and show that it has several desirable properties.