Research map / Learning theory and algorithms / Clustering
t-SNE Exaggerates Clusters, Provably
Noah Bergam, Szymon Snoeck, Nakul Verma
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add a single poison point at the data mean | t-SNE · two-Gaussian mixture, 400 points in R^2000 | cluster visibility in embedding: 400 points -> 401 points, clusters destroyed | Hurt |
| inject about 10% poison points (220) | t-SNE · BBC news RoBERTa embeddings (n=2225, D=1024) | silhouette score: halved silhouette score with 220 poison points | Hurt |
| inject about 50% outliers (1100, alpha in (0.8,1)) | t-SNE · BBC news RoBERTa embeddings | silhouette score: 1100 outliers, no degradation | No clear effect |
| apply PCA instead of t-SNE on outlier data | t-SNE · fraud data (n=5050, D=30) | outlier visibility | Improved |
| PCA pre-processing before t-SNE (ablation) | t-SNE · mixture of two Gaussians (1000 points, 100 dims) | additive invariance | Mixed |
Closest papers by meaning
- Theoretical Foundations of t-SNE for Visualizing High-Dimensional Clustered Data (JMLR 2022 · similarity 0.92)
- From $t$-SNE to UMAP with contrastive learning (ICLR 2023 · similarity 0.88)
- Your Contrastive Learning Is Secretly Doing Stochastic Neighbor Embedding (ICLR 2023 · similarity 0.86)
- WatE: A Wasserstein t-distributed Embedding Method for Information-enriched Graph Visualization (AAAI 2025 · similarity 0.86)
- THESAURUS: Contrastive Graph Clustering by Swapping Fused Gromov-Wasserstein Couplings (AAAI 2025 · similarity 0.85)
- When Edge Independence Fails: Joint Graph Diffusion with Latent Sociability Priors (NeurIPS 2026 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
