Supervised Stochastic Neighbor Embedding Using Contrastive Learning

TL;DR

This paper introduces a supervised contrastive learning approach for stochastic neighbor embedding, enhancing data visualization by effectively leveraging label information to improve class separation in low-dimensional space.

Contribution

It extends self-supervised contrastive learning to a fully-supervised setting for SNE, improving class clustering and separation in data visualization.

Findings

Enhanced class clustering in low-dimensional embeddings

Effective utilization of label information in SNE

Improved visualization quality for labeled data

Abstract

Stochastic neighbor embedding (SNE) methods $t$-SNE, UMAP are two most popular dimensionality reduction methods for data visualization. Contrastive learning, especially self-supervised contrastive learning (SSCL), has showed great success in embedding features from unlabeled data. The conceptual connection between SNE and SSCL has been exploited. In this work, within the scope of preserving neighboring information of a dataset, we extend the self-supervised contrastive approach to the fully-supervised setting, allowing us to effectively leverage label information. Clusters of samples belonging to the same class are pulled together in low-dimensional embedding space, while simultaneously pushing apart clusters of samples from different classes.