The Nearest-Instance-Centroid-Estimation Kernel Recursive Least Squares Algorithms

The nearest-instance-centroid-estimation kernel least mean-square (NICE-KLMS) algorithm has been proposed to balance the time and space requirements in kernel adaptive filters. However, the minimum mean square error (MMSE) criterion used in NICE-KLMS leads to performance degradation in some nonlinea...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE transactions on circuits and systems. II, Express briefs Jg. 67; H. 7; S. 1344 - 1348
Hauptverfasser: Zhang, Haonan, Wang, Lin, Zhang, Tao, Wang, Shiyuan
Format: Journal Article
Sprache:Englisch
Veröffentlicht: New York IEEE 01.07.2020
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Schlagworte:
Adaptive filters
Algorithms
Centroids
Circuits and systems
Clustering algorithms
Computer simulation
Dictionaries
Kernel
kernel recursive least squares algorithm
Kernels
Least squares
Mean square error methods
minimum mean square error
Nearest-instance-centroid-estimation
nonlinear problems
Performance degradation
Prediction algorithms
quantization
Signal processing algorithms
Vector quantization
ISSN:1549-7747, 1558-3791
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The nearest-instance-centroid-estimation kernel least mean-square (NICE-KLMS) algorithm has been proposed to balance the time and space requirements in kernel adaptive filters. However, the minimum mean square error (MMSE) criterion used in NICE-KLMS leads to performance degradation in some nonlinear problems. In this brief, the NICE is developed under the least-squares errors in the kernel space, generating a novel NICE kernel recursive least squares (NICE-KRLS) algorithm for performance improvement of NICE-KLMS. The weight update form for the solution to the least-squares errors existing in NICE-KRLS is therefore obtained recursively. To obtain a sparsification network, the vector quantization is combined into NICE-KRLS for online applications. Simulations on chaotic time-series prediction validate the superiorities of the proposed NICE-KRLS and its sparsification version.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1549-7747
1558-3791
DOI:10.1109/TCSII.2019.2933849