We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math.PR

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Mathematics > Probability

Title: Products of Many Large Random Matrices and Gradients in Deep Neural Networks

Abstract: We study products of random matrices in the regime where the number of terms and the size of the matrices simultaneously tend to infinity. Our main theorem is that the logarithm of the $\ell_2$ norm of such a product applied to any fixed vector is asymptotically Gaussian. The fluctuations we find can be thought of as a finite temperature correction to the limit in which first the size and then the number of matrices tend to infinity. Depending on the scaling limit considered, the mean and variance of the limiting Gaussian depend only on either the first two or the first four moments of the measure from which matrix entries are drawn. We also obtain explicit error bounds on the moments of the norm and the Kolmogorov-Smirnov distance to a Gaussian. Finally, we apply our result to obtain precise information about the stability of gradients in randomly initialized deep neural networks with ReLU activations. This provides a quantitative measure of the extent to which the exploding and vanishing gradient problem occurs in a fully connected neural network with ReLU activations and a given architecture.
Comments: v1. 26p. Comments Welcome
Subjects: Probability (math.PR); Mathematical Physics (math-ph); Machine Learning (stat.ML)
DOI: 10.1007/s00220-019-03624-z
Cite as: arXiv:1812.05994 [math.PR]
  (or arXiv:1812.05994v1 [math.PR] for this version)

Submission history

From: Boris Hanin [view email]
[v1] Fri, 14 Dec 2018 15:59:34 GMT (35kb)

Link back to: arXiv, form interface, contact.