ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2308.03812
113
7
v1v2 (latest)

Noncompact uniform universal approximation

7 August 2023
T. V. Nuland
ArXiv (abs)PDFHTML
Abstract

The universal approximation theorem is generalised to uniform convergence on the (noncompact) input space Rn\mathbb{R}^nRn. All continuous functions that vanish at infinity can be uniformly approximated by neural networks with one hidden layer, for all activation functions φ\varphiφ that are continuous, nonpolynomial, and asymptotically polynomial at ±∞\pm\infty±∞. When φ\varphiφ is moreover bounded, we exactly determine which functions can be uniformly approximated by neural networks, with the following unexpected results. Let Nφl(Rn)‾\overline{\mathcal{N}_\varphi^l(\mathbb{R}^n)}Nφl​(Rn)​ denote the vector space of functions that are uniformly approximable by neural networks with lll hidden layers and nnn inputs. For all nnn and all l≥2l\geq2l≥2, Nφl(Rn)‾\overline{\mathcal{N}_\varphi^l(\mathbb{R}^n)}Nφl​(Rn)​ turns out to be an algebra under the pointwise product. If the left limit of φ\varphiφ differs from its right limit (for instance, when φ\varphiφ is sigmoidal) the algebra Nφl(Rn)‾\overline{\mathcal{N}_\varphi^l(\mathbb{R}^n)}Nφl​(Rn)​ (l≥2l\geq2l≥2) is independent of φ\varphiφ and lll, and equals the closed span of products of sigmoids composed with one-dimensional projections. If the left limit of φ\varphiφ equals its right limit, Nφl(Rn)‾\overline{\mathcal{N}_\varphi^l(\mathbb{R}^n)}Nφl​(Rn)​ (l≥1l\geq1l≥1) equals the (real part of the) commutative resolvent algebra, a C*-algebra which is used in mathematical approaches to quantum theory. In the latter case, the algebra is independent of l≥1l\geq1l≥1, whereas in the former case Nφ2(Rn)‾\overline{\mathcal{N}_\varphi^2(\mathbb{R}^n)}Nφ2​(Rn)​ is strictly bigger than Nφ1(Rn)‾\overline{\mathcal{N}_\varphi^1(\mathbb{R}^n)}Nφ1​(Rn)​.

View on arXiv
Comments on this paper