Research map / Learning theory and algorithms / Prediction and decision losses
On UMAP's True Loss Function
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use the diverged layout instead of the UMAP result | UMAP · Toy ring, usual input similarities | purported UMAP loss: 70235 ± 1301 -> 62959 ± 82 | Helped |
| Initialize with the original layout and use dense input similarities (ablation) | UMAP · Toy ring, dense input similarities | purported UMAP loss: 224584 ± 8104 -> 331666± 1308 | No clear effect |
| Invert the positive high-dimensional similarity weights (ablation) | UMAP · C. elegans | visualization quality | Helped |
| Replace positive similarities with binary shared-kNN graph weights (ablation) | UMAP · C. elegans | visualization structure | No clear effect |
| Randomly permute positive high-dimensional similarity weights (ablation) | UMAP · C. elegans | visualization structure | No clear effect |
| Replace positive high-dimensional similarities with uniform random weights (ablation) | UMAP · C. elegans | visualization structure | No clear effect |
| Add repulsion updates for the negative-sample tail (ablation) | UMAP · Toy ring | visualization appearance | No clear effect |
| Use dense input similarities computed with the embedding similarity function (ablation) | UMAP · Toy ring | ring width | Mixed |
Closest papers by meaning
- From $t$-SNE to UMAP with contrastive learning (ICLR 2023 · similarity 0.88)
- UEQMS: UMAP Embedded Quick Mean Shift Algorithm for High Dimensional Clustering (AAAI 2023 · similarity 0.84)
- Optimization Can Learn Johnson Lindenstrauss Embeddings (NeurIPS 2024 · similarity 0.82)
- A Generic Family of Graphical Models: Diversity, Efficiency, and Heterogeneity (ICML 2025 · similarity 0.82)
- On the Recoverability of Causal Relations from Bulk Gene Expression Data (NeurIPS 2026 · similarity 0.82)
- Contrastive losses as generalized models of global epistasis (NeurIPS 2024 · similarity 0.82)
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.