A Dimensionality Reduction Method for Data Visualization using Particle Swarm Optimization

Panagiotis Petrantonakis, Ioannis Kompatsiaris


Dimensionality reduction involves mapping of a set of high dimensional input points on a low dimensional space. Mappings in low dimensional space are expected to preserve the pairwise distances of the high dimensional inputs. In this work we present a dimensionality reduction method, called Dimensionality Reduction based on Particle Swarm Optimization (PSO-DR), where the conversion of each input to the low dimensional output does not depend on the rest of the inputs but, instead, it is based on a set of reference points (beacons). The presented approach results in a simple, fast, versatile dimensionality reduction approach with good quality of visualization and straightforward out-of-sample extension.


Paper Citation