The Manifold-Index Model: provably and efficiently avoiding the statistical and computational curse of dimensionality
Organizations: Department of Mathematics, Johns Hopkins University, Baltimore, MD 21218, USA · Department of Mathematics and Department of Applied Mathematics & Statistics, Johns Hopkins University, Baltimore, MD 21218, USA
Abstract
We study vector-valued regression with predictors filling a tube around an unknown -dimensional manifold in . In the Manifold-Index Model, the mean response is , where is the nearest-point projection onto and is a possibly non-injective link function. Multiscale ANisotropic lOcal Regression (MANOR) estimates tangent planes by multiscale PCA of predictors with similar responses, then refines them by pooling outer products of local slopes. The neighborhoods narrow tangentially while retaining complete normal fibers. We assume a Lipschitz link differential and bounded response noise. For , under our geometric and sampling conditions, independent observations give mean tangent-space error of order and integrated squared prediction error at the -dimensional nonparametric rate , up to logarithmic factors. In a scaling regime with fixed intrinsic dimension, model and scale parameters, and ratio of manifold volume to the -th power of its reach, the required sample size grows polynomially in , and computation is polynomial in , and . Numerical experiments on curves with non-monotone links and surfaces with vector-valued responses compare observed convergence rates with the theoretical bounds; a Langevin dynamics example tests prediction of hitting probabilities under an approximate manifold-index model.
Figures & tables
| Symbol | Meaning in the main text |
|---|---|
| Phase 2 tangential width, (Section 2.4 ). | |
| Prediction tangential width, (Section 2.5 ). | |
| Tube-to-reach ratio, (Assumption (A1) ). | |
| Constant in the lower bound ( 3.5 ). | |
| Pooled outer-product matrix whose leading eigenspace gives (Section 2.4 ). | |
| Geodesic distance on (Section 1.5 ). |
| experiment | what it tests | Section | |||
|---|---|---|---|---|---|
| single-index (segment) | 1 | 18, 24, 30 | 1 | rates for a flat manifold | 4.2 |
| Meyer helix | 1 | 18, 24, 30 | 1 | rates for a curve of growing complexity | 4.3 |
| bumpy sphere | 2 | 20 | 6 | rates for a curved surface | 4.4 |
| coupled Meyer helix | 2 | 20 | 3 | rates for an oscillatory surface | 4.5 |
| first-hit probabilities | 2 | 20 | 4 | robustness to an approximate model | 4.6 |
| link function | error | (six points) | (five points) | reference |
|---|---|---|---|---|
| relative mean squared error | ||||
| tangent-space error | ||||
| relative mean squared error | ||||
| tangent-space error |
Appendix figures & tables5 assets
Supplementary material from the paper’s appendix.
Appendix
| link function | |||||||
|---|---|---|---|---|---|---|---|
| link function | |||||||
|---|---|---|---|---|---|---|---|
Nonparametric Regression
Nonparametric regression estimates flexible relationships between predictors and outcomes without fixing a finite-dimensional model in advance.
See all 56 papers