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


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


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

Computer Science > Machine Learning

Title: Stability of Deep Neural Networks via discrete rough paths

Abstract: Using rough path techniques, we provide a priori estimates for the output of Deep Residual Neural Networks in terms of both the input data and the (trained) network weights. As trained network weights are typically very rough when seen as functions of the layer, we propose to derive stability bounds in terms of the total $p$-variation of trained weights for any $p\in[1,3]$. Unlike the $C^1$-theory underlying the neural ODE literature, our estimates remain bounded even in the limiting case of weights behaving like Brownian motions, as suggested in [arXiv:2105.12245]. Mathematically, we interpret residual neural network as solutions to (rough) difference equations, and analyse them based on recent results of discrete time signatures and rough path theory.
Comments: 21 pages, 2 figures
Subjects: Machine Learning (cs.LG); Classical Analysis and ODEs (math.CA); Numerical Analysis (math.NA)
MSC classes: 60L90 (Primary), 39A30, 68T07 (Secondary)
Cite as: arXiv:2201.07566 [cs.LG]
  (or arXiv:2201.07566v1 [cs.LG] for this version)

Submission history

From: Nikolas Tapia [view email]
[v1] Wed, 19 Jan 2022 12:40:28 GMT (705kb,D)

Link back to: arXiv, form interface, contact.