Get Started
Home
Topics
Search
Library
Research questionHow can t-SNE’s non-convex energy produce many critical points that break topology and create spurious clusters?Because t-SNE optimizes a non-convex energy, stationary configurations may fail to preserve the data’s topology or genuine clustering structure. The difficulty is explaining why these failure modes can occur systematically rather than as isolated numerical accidents.
AI
Machine Learning
Research Paper
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.On the Abundance of Critical Points of the t-SNE EnergyThe analysis covers the original t-SNE energy and related large-data-limit energies for feature-space densities with continuous symmetry. It provides analytical and numerical examples of infinite families of distinct critical points, including configurations exhibiting topology breaking and spurious clustering.research paper · Sep 3, 2026
Related questions
How can local loss-landscape geometry reveal algorithmic structure in learned Turing-machine computations?How should distance be defined for high-dimensional clustering when Euclidean geometry mishandles anisotropy and feature interactions?How can graph neural networks capture long-range interactions without oversmoothing or oversquashing on diverse large graphs?How can we learn sparse symmetric similarity graphs that are doubly stochastic and enforce a chosen number of clusters?