Robust Kernel Adaptive Learning for Online Supervised Systems

MISSING IMAGE

Material Information

Title:
Robust Kernel Adaptive Learning for Online Supervised Systems
Physical Description:
1 online resource (128 p.)
Language:
english
Creator:
Pokharel, Rosha
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Electrical and Computer Engineering
Committee Chair:
PRINCIPE,JOSE C
Committee Co-Chair:
RANGARAJAN,ANAND
Committee Members:
HARRIS,JOHN GREGORY
CHEN,YUNMEI

Subjects

Subjects / Keywords:
adaptive -- classification -- correntropy -- filtering -- kernel -- mixture-kernel -- multi-kernel -- online -- rkhs
Electrical and Computer Engineering -- Dissertations, Academic -- UF
Genre:
Electrical and Computer Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
Kernel methods are popular for their unique ability to solve non-linear problems linearly by the implicit mapping of data into high dimension feature space. They have therefore, been established as powerful methods for solving machine learning problems. However, a method of kernel selection for these methods, specifically online kernel methods, still remains as an important problem without a widely accepted solution. Selecting an appropriate kernel can significantly improve the performance of an online learning system in small data sets but it is crucial in non-stationary environments. This dissertation proposes a unifying treatment of both filtering and classification using the same model, learning algorithms and cost functions. It presents a novel online kernel selection method based on kernel least means square (KLMS) filtering, called mixture kernel least means square (MxKLMS). MxKLMS competitively combines hypotheses learned in multiple reproducing kernel Hilbert spaces (RKHSs) such that the final hypothesis exists in the sum of these spaces. This architecture allows an automatic online kernel(s) selection from a pool of predefined kernels, such that the learning system can adapt to changing properties of the signal by changing the kernels. The detail theoretical analysis corroborates its formulation and empirical findings. Moreover, by using a mechanism of tactically discarding irrelevant samples, its memory requirement and computational burden are reduced. We compare the performance of MxKLMS trained with the conventional MSE criterion, with correntropy and show better performance in presence of impulsive noise. Apart from filtering, another major topic of supervised learning is classification. Using the same correntropy criterion on the MxKLMS model, we show its advantages in classification. Therefore, an information theoretic learning based loss function called correntropy loss (c loss) is used in the KLMS framework to obtain a robust online classifier called kernel adaptive classifier (KAC). This is extended to the MxKLMS framework to get a robust online classifier, called mixture kernel adaptive classifier (MxKAC) and is tested in non-stationary data classification. Thus, we obtain a robust method suitable for online supervised learning that includes filtering as well as classification.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Rosha Pokharel.
Thesis:
Thesis (Ph.D.)--University of Florida, 2014.
Local:
Adviser: PRINCIPE,JOSE C.
Local:
Co-adviser: RANGARAJAN,ANAND.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-05-31

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2014
System ID:
UFE0046621:00001