English

Eigenvalues of Autoencoders in Training and at Initialization

Machine Learning 2022-01-31 v1

Abstract

In this paper, we investigate the evolution of autoencoders near their initialization. In particular, we study the distribution of the eigenvalues of the Jacobian matrices of autoencoders early in the training process, training on the MNIST data set. We find that autoencoders that have not been trained have eigenvalue distributions that are qualitatively different from those which have been trained for a long time (>>100 epochs). Additionally, we find that even at early epochs, these eigenvalue distributions rapidly become qualitatively similar to those of the fully trained autoencoders. We also compare the eigenvalues at initialization to pertinent theoretical work on the eigenvalues of random matrices and the products of such matrices.

Cite

@article{arxiv.2201.11813,
  title  = {Eigenvalues of Autoencoders in Training and at Initialization},
  author = {Benjamin Dees and Susama Agarwala and Corey Lowman},
  journal= {arXiv preprint arXiv:2201.11813},
  year   = {2022}
}

Comments

11 pages, 3 figures

R2 v1 2026-06-24T09:06:19.394Z