Dimensionality Reduction Paper

Link to paper

Accompanying Repo page

Abstract: This paper examines PCA as well as non-linear reduction algorithms before delving into “Fenchel Game No Regret Dynamics” - a framework for converting convex function minimization into min-max style games - and applying one such method to a reformulation of PCA.

Along the way, this paper proposes an improvement to Laplacian Eigenmaps (non-linear reduction algorithm) to maintain structure despite oversampling, as visualized below.

Oversampled Swiss Roll

Standard LE

Proposed "Variable Radius" LE