Research map / Probabilistic and scientific modeling / Variational and generative sampling
Intrinsic dimensionality estimation using Normalizing Flows
Christian Horvat, Jean-Pascal Pfister
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 |
|---|---|---|---|
| use ID-NF instead of LIDL on correlated-on-sphere data | ID-NF · Synthetic sphere | estimated ID (true 2): 1.66±0.07 -> 2.04 | Improved |
| use ID-NF instead of LIDL on correlated swiss roll | ID-NF · Synthetic swiss roll | estimated ID (true 2): 2.48±0.03 -> 2.02 | Improved |
| use ID-NF instead of LIDL on mixture swiss roll | ID-NF · Synthetic swiss roll | estimated ID (true 2): 2.26±0.03 -> 2.02 | Improved |
| use ID-NF instead of twoNN on low-dimensional synthetic data | ID-NF · Synthetic manifolds D<=4 | estimated ID: 1.98 vs 2.01 (mixture on sphere) | No clear effect |
| use ID-NF instead of twoNN in high embedding dimension (D up to 400) | ID-NF · Sphere S(D/2) in R^D, 10^4 samples | estimated ID | Improved |
| reduce training set from 10^4 to 10^3 samples, ID-NF vs LIDL | ID-NF · Sphere S(D/2) in R^D, 10^3 samples | estimated ID | Improved |
| use a single shared sigma^2 range for LIDL across distributions (no per-distribution tuning) | LIDL · Synthetic manifolds | estimated ID: 1.66±0.07 on correlated sphere | Hurt |
| vary LIDL inflation-noise range on StyleGan images | LIDL · StyleGan2 d=2/d=64 | estimated ID | Hurt |
| use sigma_max = 0.68*255 and only 3 trained flows on images | ID-NF · StyleGan2 d=2 | estimated ID (true 2): 4.06 | Mixed |
| use sigma_max = 0.68*255 and only 3 trained flows on images | ID-NF · StyleGan2 d=64 | estimated ID (true 64): 62.24 | Improved |
| train DNF with 130 latent dims from ID-NF instead of 512 | DNF · CelebA-HQ | FID: 34.14 -> 36.92 | Mixed |
| compare DNF with 130 latent dims against M-flow with 512 | DNF (130 dims) · CelebA-HQ | FID: 38.07 -> 36.92 | Improved |
Closest papers by meaning
- A Universal Nearest-Neighbor Estimator for Intrinsic Dimensionality (TPAMI 2026 · similarity 0.88)
- Denoising Normalizing Flow (NeurIPS 2021 · similarity 0.87)
- Diffusion Models Encode the Intrinsic Dimension of Data Manifolds (ICML 2024 · similarity 0.84)
- 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 0.84)
- Learning Distances from Data with Normalizing Flows and Score Matching (ICML 2025 · 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.
