Dimensionality reduction techniques play a pivotal role in transforming high-dimensional data into lower-dimensional representations while preserving essential information. In metric learning, where the emphasis lies on capturing the underlying structure of the data space, dimensionality reduction methods are crucial for effective data classification. This paper proposes Supervised K-ISOMAP (SK-ISOMAP) aimed at integrating the principles of metric learning with the capabilities of manifold learning. By incorporating class labels or similarity constraints, SK-ISOMAP learns a discriminative embedding that not only captures the underlying structure of the data but also respects the inherent class relationships. Experimental evaluations conducted on various benchmark datasets demonstrate the efficacy of the proposed SK-ISOMAP method compared to state-of-the-art dimensionality reduction techniques, such as Supervised Uniform Manifold Approximation (Supervised UMAP). Our results highlight the ability of SK-ISOMAP to generate compact and discriminative representations, which significantly enhance the performance of supervised classifiers, using less parameters than Supervised UMAP.
alexandrelevada/SK-ISOMAP
Supervised K-ISOMAP for dimensionality reduction based metric learning
Python