Unsupervised learning aims to capture the underlying structure of potentially large and high-dimensional datasets. Traditionally, this involves using dimensionality reduction methods to project data onto lower-dimensional spaces or organizing points into meaningful clusters (clustering). Typically, this process involves aligning two graphs depicting the relationship between samples in the input high-dimensional space and their corresponding positions in the output low-dimensional space. In this talk we will present a new perspective on these approaches that is based on optimal transport and the Gromov-Wasserstein distance. Precisely, we will propose a new general framework, called distributional reduction, that recovers dimension reduction and clustering as special cases and allows us to address them jointly with a single optimization problem. We then empirically showcase the relevance of our approach on both image and genomics datasets.