Pith. sign in

REVIEW 6 cited by

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2602.10680 v1 pith:HFRU4635 submitted 2026-02-11 stat.ML cond-mat.dis-nncs.LG

A solvable high-dimensional model where nonlinear autoencoders learn structure invisible to PCA while test loss misaligns with generalization

classification stat.ML cond-mat.dis-nncs.LG
keywords modelnonlinearstructureautoencodershigh-dimensionallatentlinearloss
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

Many real-world datasets contain hidden structure that cannot be detected by simple linear correlations between input features. For example, latent factors may influence the data in a coordinated way, even though their effect is invisible to covariance-based methods such as PCA. In practice, nonlinear neural networks often succeed in extracting such hidden structure in unsupervised and self-supervised learning. However, constructing a minimal high-dimensional model where this advantage can be rigorously analyzed has remained an open theoretical challenge. We introduce a tractable high-dimensional spiked model with two latent factors: one visible to covariance, and one statistically dependent yet uncorrelated, appearing only in higher-order moments. PCA and linear autoencoders fail to recover the latter, while a minimal nonlinear autoencoder provably extracts both. We analyze both the population risk, and empirical risk minimization. Our model also provides a tractable example where self-supervised test loss is poorly aligned with representation quality: nonlinear autoencoders recover latent structure that linear methods miss, even though their reconstruction loss is higher.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. High-Dimensional Theory of LoRA Fine-Tuning in a Solvable Attention Model

    cs.LG 2026-06 unverdicted novelty 7.0

    In a solvable attention model, pre-training followed by rank-one LoRA admits sharp asymptotic predictions for test errors and representation alignment via an effective noise term.

  2. The Interplay of Data Structure and Imbalance in the Learning Dynamics of Diffusion Models

    stat.ML 2026-05 unverdicted novelty 7.0

    Higher-variance classes are learned first in diffusion models; strong class imbalance reverses the order and imposes distinct delayed learning times on minority classes.

  3. Signal-to-Noise Ratio and Sample Size Govern Representational Alignment in Neural Networks

    stat.ML 2026-05 unverdicted novelty 6.0

    Representational alignment varies monotonically with SNR and non-monotonically with sample size (minimized near interpolation threshold) across linear and nonlinear networks, and is decoupled from generalization error.

  4. Memorisation, convergence and generalisation in generative models

    stat.ML 2026-05 unverdicted novelty 6.0

    Linear generative models memorize at small data loads but converge continuously once samples scale linearly with dimension; this convergence is insensitive to sharp recovery of principal latent factors.

  5. A Fourier perspective on the learning dynamics of neural networks: from sample complexities to mechanistic insights

    stat.ML 2026-05 conditional novelty 6.0

    Neural networks prioritize amplitude over phase in Fourier space during training on translation-invariant data; power-law spectra accelerate phase learning despite not aiding classification.

  6. A theory of learning data statistics in diffusion models, from easy to hard

    stat.ML 2026-03 unverdicted novelty 6.0

    Diffusion models exhibit a distributional simplicity bias, learning pairwise input statistics at linear sample complexity while fourth-order cumulants require cubic complexity unless sharing correlated latent structure.