The fat-shattering dimension characterizes the uniform convergence property of real-valued function classes. The state-of-the-art upper bounds in Bartlett and Long (1995) feature a multiplicative squared logarithmic factor on the sample complexity, leaving an open gap with the existing lower bound. By relying on a refined packing number bound given in Rudelson and Vershynin (2006), we provide an improved uniform convergence bound that closes this gap.

An improved uniform convergence bound with fat-shattering dimension / R. Colomboni, E. Esposito, A. Paudice. - In: INFORMATION PROCESSING LETTERS. - ISSN 0020-0190. - 188:(2025 Feb), pp. 106539.1-106539.6. [10.1016/j.ipl.2024.106539]

An improved uniform convergence bound with fat-shattering dimension

R. Colomboni
Primo
;
E. Esposito
Penultimo
;
A. Paudice
Ultimo
2025

Abstract

The fat-shattering dimension characterizes the uniform convergence property of real-valued function classes. The state-of-the-art upper bounds in Bartlett and Long (1995) feature a multiplicative squared logarithmic factor on the sample complexity, leaving an open gap with the existing lower bound. By relying on a refined packing number bound given in Rudelson and Vershynin (2006), we provide an improved uniform convergence bound that closes this gap.
Uniform convergence; Fat-shattering dimension; Chaining;
Settore INFO-01/A - Informatica
   Learning in Markets and Society
   MINISTERO DELL'UNIVERSITA' E DELLA RICERCA
   2022EKNE5K_001

   European Lighthouse of AI for Sustainability (ELIAS)
   ELIAS
   EUROPEAN COMMISSION
   101120237
feb-2025
nov-2024
https://www.sciencedirect.com/science/article/pii/S0020019024000693
Article (author)
File in questo prodotto:
File Dimensione Formato  
1-s2.0-S0020019024000693-main.pdf

accesso riservato

Tipologia: Publisher's version/PDF
Licenza: Nessuna licenza
Dimensione 490.98 kB
Formato Adobe PDF
490.98 kB Adobe PDF   Visualizza/Apri   Richiedi una copia
preprint.pdf

accesso aperto

Tipologia: Pre-print (manoscritto inviato all'editore)
Licenza: Creative commons
Dimensione 197.73 kB
Formato Adobe PDF
197.73 kB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/2434/1206856
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? 0
  • OpenAlex ND
social impact