Cluster Representatives Selection in Non-Metric Spaces for Nearest Prototype Classification

TL;DR

This paper introduces CRS, a new graph-based method for selecting representative prototypes in non-metric spaces, improving efficiency and accuracy in nearest prototype classification across diverse datasets.

Contribution

CRS is a novel, efficient prototype selection method that works in arbitrary metric and non-metric spaces using similarity graphs, outperforming existing techniques.

Findings

CRS outperforms state-of-the-art methods on multiple datasets.

The graph-based approach enables efficient prototype selection in non-metric spaces.

CRS is applicable in various domains due to its reliance on pairwise similarity measures.

Abstract

The nearest prototype classification is a less computationally intensive replacement for the $k$-NN method, especially when large datasets are considered. In metric spaces, centroids are often used as prototypes to represent whole clusters. The selection of cluster prototypes in non-metric spaces is more challenging as the idea of computing centroids is not directly applicable. In this paper, we present CRS, a novel method for selecting a small yet representative subset of objects as a cluster prototype. Memory and computationally efficient selection of representatives is enabled by leveraging the similarity graph representation of each cluster created by the NN-Descent algorithm. CRS can be used in an arbitrary metric or non-metric space because of the graph-based approach, which requires only a pairwise similarity measure. As we demonstrate in the experimental evaluation, our method outperforms the state of the art techniques on multiple datasets from different domains.