Existence and Uniqueness of Minimization Problems with Fourier Based Stabilizers
1.
SYSNO ASEP
0105194
Document Type
C - Proceedings Paper (int. conf.)
R&D Document Type
Conference Paper
Title
Existence and Uniqueness of Minimization Problems with Fourier Based Stabilizers
Title
Existence a jednoznačnost milimalizačních problémů s Fourierovským stabilizátorem
Author(s)
Šidlofová, Terezie (UIVT-O)
Source Title
COMPSTAT Proceedings in Computational Statistics. - Heidelberg : Physica-Verlag, 2004 / Antoch J.
- ISBN 978-3-7908-1554-2
Pages
s. 1853-1860
Number of pages
8 s.
Publication form
CD ROM - CD ROM
Action
COMPSTAT 2004. Symposium /16./
Event date
23.08.2004-27.08.2004
VEvent location
Prague
Country
CZ - Czech Republic
Event type
WRD
Language
eng - English
Country
DE - Germany
Keywords
neural networks ; minimization of functionals ; regularization theory ; stabilizers ; Fourier transform
Subject RIV
BA - General Mathematics
R&D Projects
GA201/02/0428 GA ČR - Czech Science Foundation (CSF)
CEZ
AV0Z1030915 - UIVT-O
Annotation
We study minimization of regularized empirical error functional with a Fourier-based stabilizer. We prove existence and uniqueness of the solution. We also describe the shape of the minimizing function and show that it is in the form of a one-hidden layer feed-forward neural network with activation functions derived from the regularization part. Practical applications based on the idea have been studied performing best on tasks with lower input dimension or suitable conceptual characteristics (e.g. financial fields or image classification).
