Research map / Learning theory and algorithms / Sample complexity
A Universal Nearest-Neighbor Estimator for Intrinsic Dimensionality
Eng-Jon Ong, Omer Bobrowski, Gesine Reinert, Primož Škraba
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 parameter optimization on the benchmark (alpha, beta tuned) to L2N2(2,1) (ablation) | L2N2(2,1) · Benchmark manifolds (2,500 pts) | MPE (%): 9.96 -> 8.09 | Improved |
| round estimated dimensions to nearest integer (ablation) | L2N2(2,1) · Benchmark manifolds (2,500 pts) | MPE (%): 9.96 -> 6.06 | Improved |
| use larger neighbor indices (k,j)=(8,4) instead of (2,1) (ablation) | L2N2 · Benchmark manifolds (2,500 pts) | MPE (%): 9.96 -> 12.96 | Hurt |
| use (4,2) instead of (2,1) (ablation) | L2N2 · Benchmark manifolds (2,500 pts) | MPE (%): 9.96 -> 11.38 | Hurt |
| use L2N2(2,1) instead of TwoNN | TwoNN · Benchmark manifolds (2,500 pts) | MPE (%): 10.99 -> 9.96 | Improved |
| use L2N2(2,1) instead of MLE | MLE · Benchmark manifolds (2,500 pts) | MPE (%): 15.96 -> 9.96 | Improved |
| increase sample size from 625 to 5,000 points | L2N2(2,1) · Benchmark manifolds | MPE (%): 13.02 -> 9.28 | Improved |
| subsample query points to 2,500 for large datasets (ablation) | L2N2(2,1) · Benchmark manifolds (500k pts) | MPE (%): 6.41 -> 6.88 | Hurt |
| add Gaussian ambient noise sigma 0.1 to S^6 | L2N2(2,1) · S^6 in R^11 | estimated dimension: 6.10 -> 8.48 | Hurt |
| use L2N2 estimate as autoencoder bottleneck instead of TwoNN | autoencoder · MNIST | reconstruction MSE: 0.0143 -> 0.0119 | Improved |
| use dimension range 10-40 instead of 1-20 when fitting alpha at n=2,500 (ablation) | L2N2 tuning · Gaussian calibration | alpha_{2,1}: 0.929932 -> 0.762282 | Hurt |
Closest papers by meaning
- Intrinsic dimensionality estimation using Normalizing Flows (NeurIPS 2022 · similarity 0.88)
- A Wiener Process Perspective on Local Intrinsic Dimension Estimation Methods (AAAI 2025 · similarity 0.84)
- Intrinsic Dimension Correlation: uncovering nonlinear connections in multimodal representations (ICLR 2025 · similarity 0.83)
- Nearest-Neighbor Radii under Dependent Sampling (NeurIPS 2026 · similarity 0.83)
- Estimating Shape Distances on Neural Representations with Limited Samples (ICLR 2024 · similarity 0.83)
- How to Overcome Curse-of-Dimensionality for Out-of-Distribution Detection? (AAAI 2024 · 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.
