In this paper, we present a novel computational framework for nonlinear dimensionality reduction which is specifically suited to process large data sets: the Exploratory Inspection Machine (XIM). XIM introduces a conceptual cross-link between hitherto separate domains of machine learning, namely topographic vector quantization and divergence-based neighbor embedding approaches. There are three ways to conceptualize XIM, namely (i) as the inversion of the Exploratory Observation Machine (XOM) and its variants, such as Neighbor Embedding XOM (NE-XOM), (ii) as a powerful optimization scheme for divergence-based neighbor embedding cost functions inspired by Stochastic Neighbor Embedding (SNE) and its variants, such as t-distributed SNE (t-SNE), and (iii) as an extension of topographic vector quantization methods, such as the Self-Organizing Map (SOM). By preserving both global and local data structure, XIM combines the virtues of classical and advanced recent embedding methods. It permits direct visualization of large data collections without the need for prior data reduction. Finally, XIM can contribute to many application domains of data analysis and visualization important throughout the sciences and engineering, such as pattern matching, constrained incremental learning, data clustering, and the analysis of non-metric dissimilarity data.