Understanding UMAP Internal Workings of a State of the Art Clustering Algorithm
summary
UMAP constructs a local neighbor graph (based on n_neighbors) and converts it into a fuzzy topological representation, then optimizes a low‑dimensional embedding by balancing attraction and repulsion forces controlled by min_dist. Demonstrates effects on real datasets — population and genomic examples — and reports how changing those parameters produces qualitatively different cluster separation, density preservation, and isolated points, with concrete plots and specific parameter values shown.
AI-written summary · about
chapters (6)
- 0:00Introduction to UMAP
- 3:59UMAP as cutting-edge clustering
- 8:00Key concepts in UMAP
- 13:05Distance and parameters
- 19:11Applications and insights
- 23:53References and conclusion
related talks
40mTopological Techniques for Unsupervised Learning
25mThe Art of Clustering: The Good, The Bad and The Beautiful
49mMetric Representations: Algorithms and Geometry
26mMathematical Approaches to Clustering
27mHow I hacked UMAP and won at a plotting contest
47mAssessing the quality of a clustering
35mClustering: A Guide for the Perplexed
50mThe Shape of Data: Machine Learning and Topology