What you'll learn

Most real-world datasets and every learned embedding space live in far too many dimensions to inspect directly, and linear tools like PCA flatten the nonlinear structure that matters most. In this course, Nonlinear Dimensionality Reduction and Visualization, you’ll gain the ability to project high-dimensional data with t-SNE and UMAP and interpret the resulting plots without fooling yourself. First, you’ll explore why linear projections fail on curved manifolds and how to recognize the dataset characteristics that call for a nonlinear method. Next, you’ll discover how t-SNE preserves local neighborhoods, how perplexity and initialization change the picture, and which parts of a t-SNE plot you are allowed to believe. Finally, you’ll learn how to tune UMAP for both visualization and downstream clustering, validate visual insights across multiple projections and parameter settings, and use these techniques to explore and debug the embedding spaces produced by real ML models. When you’re finished with this course, you’ll have the skills and knowledge of nonlinear dimensionality reduction and visualization needed to reveal genuine structure in high-dimensional data and defend your interpretation of it.