Research map / Label-efficient and robust learning / Recognition and long-tail
Attraction-Repulsion Spectrum in Neighbor Embeddings
Jan Niklas Böhm, Philipp Berens, Dmitry Kobak
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 |
|---|---|---|---|
| Increase t-SNE attraction by raising exaggeration from ρ=1 to ρ=4 | t-SNE · MNIST | kNN recall: 0.34 -> 0.12 | Hurt |
| Increase t-SNE attraction by raising exaggeration from ρ=4 to ρ=30 | t-SNE · MNIST | kNN recall: 0.12 -> 0.06 | Hurt |
| Increase t-SNE attraction | t-SNE · Simulated 20-Gaussian trajectory | continuity of the recovered manifold | Improved |
| Increase t-SNE attraction | t-SNE · Human brain organoid developmental data | representation of developmental trajectory | Improved |
| Increase t-SNE repulsion by using standard exaggeration instead of strong attraction | t-SNE · Human brain organoid developmental data | representation of cluster structure | Improved |
| Use UMAP negative sampling with larger rate m (ablation) | UMAP · MNIST subsample | repulsion strength | Improved |
| Increase negative sampling rate m while decreasing γ to keep γ·m constant (ablation) | UMAP · MNIST subsample | embedding | No clear effect |
| Decrease UMAP repulsion coefficient γ (ablation) | UMAP · MNIST | attraction-repulsion spectrum position | Improved |
| Set UMAP output-kernel parameters to a=b=1 (ablation) | UMAP · MNIST | embedding | No clear effect |
| Set UMAP ε=1 (ablation) | UMAP · MNIST | embedding | No clear effect |
| Remove degree-based repulsion (ablation) | ForceAtlas2 · MNIST | embedding scale: ∼15 times decrease in scale | Hurt |
| Remove degree-based repulsion (ablation) | ForceAtlas2 · MNIST | embedding layout | No clear effect |
Closest papers by meaning
- Embedding Dimension of Contrastive Learning and $k$-Nearest Neighbors (NeurIPS 2024 · similarity 0.83)
- Limit theorems for out-of-sample extensions of the adjacency and Laplacian spectral embeddings (JMLR 2021 · similarity 0.82)
- LoNe Sampler: Graph Node Embeddings by Coordinated Local Neighborhood Sampling (AAAI 2023 · similarity 0.82)
- Node Similarities under Random Projections: Limits and Pathological Cases (ICLR 2025 · similarity 0.79)
- Learning Shared Representations from Unpaired Data (NeurIPS 2025 · similarity 0.79)
- Weighted Embeddings for Low-Dimensional Graph Representation (AAAI 2025 · similarity 0.79)
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.
