Research map / Learning theory and algorithms / Sample complexity
Adaptive Geometric Multiscale Approximations for Intrinsically Low-dimensional Data
Wenjing Liao, Mauro Maggioni
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 adaptive GMRA instead of uniform-scale GMRA | GMRA · 4-dim Z manifold, noise-free, L2 error vs n | rate of convergence: Adaptive GMRA yields a faster rate of convergence than GMRA | Improved |
| use GMRA instead of nearest neighbor approximation | Nearest Neighbor · 4-dim S and Z manifolds, noise-free | L2 error and convergence rate: NN rate 1/d = 0.25; GMRA smaller error and faster rate | Improved |
| use larger kappa threshold (0.5 or 1) in adaptive GMRA under noise sigma = 0.05 (ablation) | Adaptive GMRA · 4-dim manifolds, sigma = 0.05 | rate of convergence: kappa = 0.5 and 1 give the fastest rate; kappa = 0.05 is similar to nearest neighbor | Improved |
| vary kappa among sufficiently large values in adaptive GMRA (ablation) | Adaptive GMRA · 4-dim manifolds, sigma = 0.05 | rate of convergence: different errors up to a constant, same rate | No clear effect |
| increase training noise level sigma (ablation) | GMRA and adaptive GMRA · 4-dim S and Z manifolds, n = 1e5 | average L2 approximation error: error increases linearly with sigma | Hurt |
| increase sample size n beyond the noise level | GMRA · 4-dim manifolds, sigma = 0.05 | L2 error vs noise level: approximation error falls below sigma as n increases | Improved |
Closest papers by meaning
- Multiscale Dictionary Learning: Non-Asymptotic Bounds and Robustness (JMLR 2016 · similarity 0.92)
- Finsler Multi-Dimensional Scaling: Manifold Learning for Asymmetric Dimensionality Reduction and Embedding (CVPR 2025 · similarity 0.86)
- The Noisy Laplacian: a Threshold Phenomenon for Non-Linear Dimension Reduction (ICML 2025 · similarity 0.85)
- Neuc-MDS: Non-Euclidean Multidimensional Scaling Through Bilinear Forms (NeurIPS 2024 · similarity 0.84)
- Sample-Efficient Geometry Reconstruction from Euclidean Distances using Non-Convex Optimization (NeurIPS 2024 · similarity 0.84)
- Manifold Random Features (NeurIPS 2026 · 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.
