Research map / Label-efficient and robust learning / Self-supervised and contrastive
Cross-modal Scalable Hyperbolic Hierarchical Clustering
Teng Long, Nanne van Noord
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 |
|---|---|---|---|
| split hyperbolic hierarchical clustering into Sinkhorn pre-clustering plus hyperbolic post-clustering (ablation) | HypHC-style clustering · Kinetics-Sound multi-modal | NMI: 20.6 -> 36.9 | Improved |
| add hyperbolic hierarchical post-clustering on top of Sinkhorn pre-clustering (ablation) | SeLaVi · Kinetics-Sound multi-modal | NMI: 36.4 -> 36.9 | Improved |
| use hyperbolic space instead of Euclidean for the hierarchy (ablation) | sHHC · Kinetics-Sound multi-modal | NMI: 36.3 -> 36.9 | Improved |
| move SeLaVi to hyperbolic space without hierarchy (ablation) | SeLaVi · Kinetics-Sound multi-modal | NMI: 36.4 -> 36.1 | No clear effect |
| use fewer clusters (34 instead of 100) (ablation) | sHHC · Kinetics-Sound | NMI: 36.9 -> 37.4 | Improved |
| use sHHC hierarchy instead of HypHC or UFit | hierarchical clustering · VGGSound hierarchy | Dasgupta cost / runtime: 7.72 -> 6.62; ~4h -> ~2min (vs UFit) | Improved |
| use sHHC hierarchical pseudo-labels for pretraining | L3 backbone · KineticsSound multi-modal recognition | accuracy: 70.95 -> 74.10 | Improved |
| use half-precision or mixed precision training | hyperbolic clustering · numerical stability | performance | Hurt |
Closest papers by meaning
- Clustering Properties of Self-Supervised Learning (ICML 2025 · similarity 0.85)
- From Logits to Hierarchies: Hierarchical Clustering made Simple (ICML 2025 · similarity 0.85)
- Classification Fields: Arbitrarily Fine Recursive Hierarchical Clustering From Few Examples (NeurIPS 2026 · similarity 0.85)
- Learning Structured Representations with Hyperbolic Embeddings (NeurIPS 2024 · similarity 0.84)
- Bringing Motion Taxonomies to Continuous Domains via GPLVM on Hyperbolic manifolds (ICML 2024 · similarity 0.84)
- Hyperbolic Category Discovery (CVPR 2025 · similarity 0.84)
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.
