Minimax rates for estimating the dimension of a manifold

DOI10.20382/JOCG.V10I1A3MaRDI QIDQ4626307zbMATH OpenOpenAlexFDO

Authors Jisu Kim, Alessandro Rinaldo, Larry Wasserman

Publication date 27 February 2019

Full work available at URL https://arxiv.org/abs/1605.01011

Mathematics Subject Classification ID

Estimation in multivariate analysis (62H12) Hypothesis testing in multivariate analysis (62H15) Probability in computer science (algorithm analysis, random structures, phase transitions, etc.) (68Q87) Learning and adaptive systems in artificial intelligence (68T05) Computer graphics; computational geometry (digital and algorithmic aspects) (68U05)

Abstract: Many algorithms in machine learning and computational geometry require, as input, the intrinsic dimension of the manifold that supports the probability distribution of the data. This parameter is rarely known and therefore has to be estimated. We characterize the statistical difficulty of this problem by deriving upper and lower bounds on the minimax rate for estimating the dimension. First, we consider the problem of testing the hypothesis that the support of the data-generating probability distribution is a well-behaved manifold of intrinsic dimension $d_1$ versus the alternative that it is of dimension $d_2$, with $d_1<d_2$. With an i.i.d. sample of size $n$, we provide an upper bound on the probability of choosing the wrong dimension of $Oleft(n^{-left(d_2/d_1-1-epsilonight)n}ight)$, where $epsilon$ is an arbitrarily small positive number. The proof is based on bounding the length of the traveling salesman path through the data points. We also demonstrate a lower bound of $Omegaleft(n^{-(2d_2-2d_1+epsilon)n}ight)$, by applying Le Cam's lemma with a specific set of $d_1$-dimensional probability distributions. We then extend these results to get minimax rates for estimating the dimension of well-behaved manifolds. We obtain an upper bound of order $Oleft(n^{-(frac1m-1-epsilon)n}ight)$ and a lower bound of order $Omegaleft(n^{-(2+epsilon)n}ight)$, where $m$ is the embedding dimension.

Recommendations

• Minimax manifold estimation
• Testing the manifold hypothesis
• Learning gradients on manifolds
• Tight minimax rates for manifold estimation under Hausdorff loss
• Minimax adaptive estimation in manifold inference

Cited in

(18)

• Density estimation on an unknown submanifold
• Minimax adaptive estimation in manifold inference
• Estimating the reach of a manifold via its convexity defect function
• Measure estimation on manifolds: an optimal transport approach
• Local angles and dimension estimation from data on manifolds
• Recovering the homology of immersed manifolds
• Determining Intrinsic Dimension and Entropy of High-Dimensional Shape Spaces
• Dimension estimation using random connection models
• scientific article; zbMATH DE number 6999890 (Why is no real title available?)
• Tight minimax rates for manifold estimation under Hausdorff loss
• scientific article; zbMATH DE number 7626706 (Why is no real title available?)
• Minimax manifold estimation
• Testing the manifold hypothesis
• Minimax rate of distribution estimation on unknown submanifolds under adversarial losses
• Statistical embedding: beyond principal components
• Tangent space and dimension estimation with the Wasserstein distance
• Nonparametric regression on random geometric graphs sampled from submanifolds
• On consistent estimation of dimension values

This page was built for publication: Minimax rates for estimating the dimension of a manifold

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4626307)