Search Dimension in Unlabeled Projection Pursuit: A Scaling Law for Subspace Restriction
Abstract
Projection pursuit searches for a direction along which the data look least Gaussian. When the observation space contains a large Gaussian complement, the empirical objective can be minimized by a direction that carries no signal, with empirical kurtosis as low as at the truth. Sample splitting exposes rather than repairs this failure. Appending coordinates independent of the latent regime degrades the search while leaving Bayes recoverability unchanged. Restricting the search to the column space of a known forward operator removes the failure exactly on the negative-kurtosis branch. Estimating a principal subspace from the data is the alternative. In a controlled two-component model, the leading sufficient scalings differ in the gain with which the operator transmits the discriminant: ς−4 for covariance-spike estimation and ς−8 for fourth-moment search. At fixed search dimension, the measured threshold ratio collapses onto n/p2 with exponent $0.156$, close to the predicted 1/8. This is an empirically supported scaling motivated by sufficient bounds, not a proved asymptotically tight law. When the search dimension is varied, the measured exponent is $0.325$, substantially larger than 1/8, and the tested range does not identify its functional form. The crossing location also depends on calibration and model configuration. Under a downstream excess-error criterion, the scaling largely disappears.
Related
- Algorithmic threshold for high-dimensional projection pursuit I: general theory
- Dimension-free regimes for unadjusted Langevin in ridge-separable models: A subspace robust Wasserstein analysis
- Tiling without a subspace complement: a sharp counterexample in dimension nine over the field with two elements
- Optimal Ferrers Diagram Rank-Metric Codes: New Constructions, Diagram Combinations, and Applications to Constant-Dimension Subspace Codes
- Fourier Restriction Beyond Dimension Data: A Multiscale Moment Criterion
BibTeX
@misc{grozavescu2026search,
title = {{Search Dimension in Unlabeled Projection Pursuit: A Scaling Law for Subspace Restriction}},
author = {Rares Dimitrie Grozavescu and Mark Girolami},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.37917},
eprint = {2609.37917},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.37917},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Search Dimension in Unlabeled Projection Pursuit: A Scaling Law for Subspace Restriction
AU - Rares Dimitrie Grozavescu
AU - Mark Girolami
PY - 2026
DA - 2026/09/29
PB - arXiv
DO - 10.48550/arxiv.2609.37917
UR - https://arxiv.org/abs/2609.37917
AB - Projection pursuit searches for a direction along which the data look least Gaussian. When the observation space contains a large Gaussian complement, the empirical objective can be minimized by a direction that carries no signal, with empirical kurtosis as low as at the truth. Sample splitting exposes rather than repairs this failure. Appending coordinates independent of the latent regime degrades the search while leaving Bayes recoverability unchanged. Restricting the search to the column space of a known forward operator removes the failure exactly on the negative-kurtosis branch. Estimating a principal subspace from the data is the alternative. In a controlled two-component model, the leading sufficient scalings differ in the gain with which the operator transmits the discriminant: ς^−4 for covariance-spike estimation and ς^−8 for fourth-moment search. At fixed search dimension, the measured threshold ratio collapses onto n/p^2 with exponent $0.156$, close to the predicted 1/8. This is an empirically supported scaling motivated by sufficient bounds, not a proved asymptotically tight law. When the search dimension is varied, the measured exponent is $0.325$, substantially larger than 1/8, and the tested range does not identify its functional form. The crossing location also depends on calibration and model configuration. Under a downstream excess-error criterion, the scaling largely disappears.
ER -