A Linear Surrogate-Based Algorithm for Fitting Gaussian Mixture Functions

Gaussian Mixture Function (GMF) is a widely utilized model for analyzing and elucidating experimental data in science and engineering, where the fitting of GMF with noisy observations is usually rendered acomplicated nonlinear regression problem due to the underlying linear superposition of Gaussian...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:IEEE signal processing letters Ročník 32; s. 3874 - 3878
Hlavní autoři: Xiao, Yucong, Li, Xuan, Dai, Xuewu, Yang, Yang, Qin, Fei
Médium: Journal Article
Jazyk:angličtina
Vydáno: New York IEEE 2025
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Témata:
Approximation algorithms
Convergence
expectation-maximization
Fitting
Gaussian mixture function
linear least squares
Lower bound
majorization-minimization
Minimization
Mixtures
Noise measurement
nonlinear regression
Posterior probability
Probability density function
Signal processing algorithms
Training
ISSN:1070-9908, 1558-2361
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Gaussian Mixture Function (GMF) is a widely utilized model for analyzing and elucidating experimental data in science and engineering, where the fitting of GMF with noisy observations is usually rendered acomplicated nonlinear regression problem due to the underlying linear superposition of Gaussian components. Classical Newton-type solutions rely on derivatives of the regression objective to facilitate convergence, which are general-purpose and can be inefficient. In this letter, we propose a novel method inspiredby Majorization-Minimization (MM) to achieve efficient GMF fitting in a linear manner. The proposed method integrates the contribution of each Gaussian component in GMF to construct a linear surrogate and ensures the consistent convergence of the original nonlinear objective. Extensive experiments demonstrate that the proposed method outperforms classical solutions in convergence speed while maintaining precise fitting accuracy.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1070-9908
1558-2361
DOI:10.1109/LSP.2025.3616610