Minimax Optimal Regression over Sobolev Spaces via Laplacian Eigenmaps on Neighborhood Graphs

TL;DR

This paper demonstrates that Principal Components Regression with Laplacian Eigenmaps (PCR-LE) achieves minimax optimal rates for nonparametric regression over Sobolev spaces, including manifold settings, outperforming traditional eigenvector convergence rates.

Contribution

The paper establishes minimax optimal convergence rates for PCR-LE in Sobolev spaces and shows its adaptivity to manifold structures, providing both theoretical guarantees and empirical validation.

Findings

PCR-LE achieves optimal rates for estimation and testing in Sobolev spaces.

PCR-LE is manifold adaptive, attaining faster rates on low-dimensional manifolds.

Regression with estimated features is statistically easier than feature estimation itself.

Abstract

In this paper we study the statistical properties of Principal Components Regression with Laplacian Eigenmaps (PCR-LE), a method for nonparametric regression based on Laplacian Eigenmaps (LE). PCR-LE works by projecting a vector of observed responses ${\bf Y} = (Y_1,\ldots,Y_n)$ onto a subspace spanned by certain eigenvectors of a neighborhood graph Laplacian. We show that PCR-LE achieves minimax rates of convergence for random design regression over Sobolev spaces. Under sufficient smoothness conditions on the design density $p$, PCR-LE achieves the optimal rates for both estimation (where the optimal rate in squared $L^2$ norm is known to be $n^{-2s/(2s + d)}$) and goodness-of-fit testing ($n^{-4s/(4s + d)}$). We also show that PCR-LE is \emph{manifold adaptive}: that is, we consider the situation where the design is supported on a manifold of small intrinsic dimension $m$, and give upper bounds establishing that PCR-LE achieves the faster minimax estimation ($n^{-2s/(2s + m)}$) and testing ($n^{-4s/(4s + m)}$) rates of convergence. Interestingly, these rates are almost always much faster than the known rates of convergence of graph Laplacian eigenvectors to their population-level limits; in other words, for this problem regression with estimated features appears to be much easier, statistically speaking, than estimating the features itself. We support these theoretical results with empirical evidence.