Residual spectral instabilities in representation learning
Abstract
Learned representations can lose latent degrees of freedom successively, suggesting a cascade of transitions whose underlying stability principle remains unclear. Here we formulate dimension-wise posterior collapse in variational autoencoder (VAE) as a fluctuation theory around partially collapsed states. Interpreting the negative evidence lower bound as an effective free energy, its quadratic expansion defines a Gaussian theory whose Hessian acts as a mass matrix for latent fluctuations. We show that the collapsed directions form an invariant fluctuation sector and derive its exact mass spectrum in terms of a conditional residual operator. A local reactivation direction lowers the free energy when the decoder variance falls below the residual spectral upper edge, with equality marking marginality. The criterion recovers principal component thresholds in the linear Gaussian VAE limit. Viewed in reverse along continuously connected branches, the reactivation boundary provides a local criterion for successive collapse. Numerical continuation experiments show successive loss of latent dimensions near these spectral marginalities. These results support a spectral cascade interpretation governed by residual information left unexplained by the surviving representation.
Related
- Compact and stable representation of real-frequency spectral functions for machine learning
- Spectral instabilities in the time domain
- Beyond de Sitter: Longitudinal-Mode Instabilities and Spectral Evolution of Massive Vector Fields with Non-Minimal Curvature Coupling during Inflation
- The Residual $288$ in an E8×ωE8 Program: Representation Accounting and Physical Matter Interfaces
- Robust acoustic representation of Japanese vowels under phonation variability using dynamic spectral analysis
BibTeX
@misc{li2026residual,
title = {{Residual spectral instabilities in representation learning}},
author = {Zhen Li},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.11257},
eprint = {2610.11257},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.11257},
}
RIS
TY - UNPB
TI - Residual spectral instabilities in representation learning
AU - Zhen Li
PY - 2026
DA - 2026/10/08
PB - arXiv
DO - 10.48550/arxiv.2610.11257
UR - https://arxiv.org/abs/2610.11257
AB - Learned representations can lose latent degrees of freedom successively, suggesting a cascade of transitions whose underlying stability principle remains unclear. Here we formulate dimension-wise posterior collapse in variational autoencoder (VAE) as a fluctuation theory around partially collapsed states. Interpreting the negative evidence lower bound as an effective free energy, its quadratic expansion defines a Gaussian theory whose Hessian acts as a mass matrix for latent fluctuations. We show that the collapsed directions form an invariant fluctuation sector and derive its exact mass spectrum in terms of a conditional residual operator. A local reactivation direction lowers the free energy when the decoder variance falls below the residual spectral upper edge, with equality marking marginality. The criterion recovers principal component thresholds in the linear Gaussian VAE limit. Viewed in reverse along continuously connected branches, the reactivation boundary provides a local criterion for successive collapse. Numerical continuation experiments show successive loss of latent dimensions near these spectral marginalities. These results support a spectral cascade interpretation governed by residual information left unexplained by the surviving representation.
ER -