Online model selection and learning by multikernel adaptive filtering

Masahiro Yukawa, Ryu Ichiro Ishii

Research output: Chapter in Book/Report/Conference proceedingConference contribution

9 Citations (Scopus)

Abstract

We propose an efficient multikernel adaptive filtering algorithm with double regularizers, providing a novel pathway towards online model selection and learning. The task is the challenging nonlinear adaptive filtering under no knowledge about a suitable kernel. Under this limited-knowledge assumption on an underlying model of a system of interest, many possible kernels are employed and one of the regularizers, a block ℓ1 norm for kernel groups, contributes to selecting a proper model (relevant kernels) in online and adaptive fashion, preventing a nonlinear filter from overfitting to noisy data. The other regularizer is the block ℓ1 norm for data groups, contributing to updating the dictionary adaptively. As the resulting cost function contains two nonsmooth (but proximable) terms, we approximate the latter regularizer by its Moreau envelope and apply the adaptive proximal forwardbackward splitting method to the approximated cost function. Numerical examples show the efficacy of the proposed algorithm.

Original languageEnglish
Title of host publicationEuropean Signal Processing Conference
PublisherEuropean Signal Processing Conference, EUSIPCO
ISBN (Print)9780992862602
Publication statusPublished - 2013
Event2013 21st European Signal Processing Conference, EUSIPCO 2013 - Marrakech, Morocco
Duration: 2013 Sep 92013 Sep 13

Other

Other2013 21st European Signal Processing Conference, EUSIPCO 2013
CountryMorocco
CityMarrakech
Period13/9/913/9/13

Fingerprint

Adaptive filtering
Cost functions
Glossaries

Keywords

  • kernel adaptive filter
  • multiple kernels
  • proximity operator

ASJC Scopus subject areas

  • Signal Processing
  • Electrical and Electronic Engineering

Cite this

Yukawa, M., & Ishii, R. I. (2013). Online model selection and learning by multikernel adaptive filtering. In European Signal Processing Conference [6811527] European Signal Processing Conference, EUSIPCO.

Online model selection and learning by multikernel adaptive filtering. / Yukawa, Masahiro; Ishii, Ryu Ichiro.

European Signal Processing Conference. European Signal Processing Conference, EUSIPCO, 2013. 6811527.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Yukawa, M & Ishii, RI 2013, Online model selection and learning by multikernel adaptive filtering. in European Signal Processing Conference., 6811527, European Signal Processing Conference, EUSIPCO, 2013 21st European Signal Processing Conference, EUSIPCO 2013, Marrakech, Morocco, 13/9/9.
Yukawa M, Ishii RI. Online model selection and learning by multikernel adaptive filtering. In European Signal Processing Conference. European Signal Processing Conference, EUSIPCO. 2013. 6811527
Yukawa, Masahiro ; Ishii, Ryu Ichiro. / Online model selection and learning by multikernel adaptive filtering. European Signal Processing Conference. European Signal Processing Conference, EUSIPCO, 2013.
@inproceedings{205c880a3b3545fbbba4b549da7b9294,
title = "Online model selection and learning by multikernel adaptive filtering",
abstract = "We propose an efficient multikernel adaptive filtering algorithm with double regularizers, providing a novel pathway towards online model selection and learning. The task is the challenging nonlinear adaptive filtering under no knowledge about a suitable kernel. Under this limited-knowledge assumption on an underlying model of a system of interest, many possible kernels are employed and one of the regularizers, a block ℓ1 norm for kernel groups, contributes to selecting a proper model (relevant kernels) in online and adaptive fashion, preventing a nonlinear filter from overfitting to noisy data. The other regularizer is the block ℓ1 norm for data groups, contributing to updating the dictionary adaptively. As the resulting cost function contains two nonsmooth (but proximable) terms, we approximate the latter regularizer by its Moreau envelope and apply the adaptive proximal forwardbackward splitting method to the approximated cost function. Numerical examples show the efficacy of the proposed algorithm.",
keywords = "kernel adaptive filter, multiple kernels, proximity operator",
author = "Masahiro Yukawa and Ishii, {Ryu Ichiro}",
year = "2013",
language = "English",
isbn = "9780992862602",
booktitle = "European Signal Processing Conference",
publisher = "European Signal Processing Conference, EUSIPCO",

}

TY - GEN

T1 - Online model selection and learning by multikernel adaptive filtering

AU - Yukawa, Masahiro

AU - Ishii, Ryu Ichiro

PY - 2013

Y1 - 2013

N2 - We propose an efficient multikernel adaptive filtering algorithm with double regularizers, providing a novel pathway towards online model selection and learning. The task is the challenging nonlinear adaptive filtering under no knowledge about a suitable kernel. Under this limited-knowledge assumption on an underlying model of a system of interest, many possible kernels are employed and one of the regularizers, a block ℓ1 norm for kernel groups, contributes to selecting a proper model (relevant kernels) in online and adaptive fashion, preventing a nonlinear filter from overfitting to noisy data. The other regularizer is the block ℓ1 norm for data groups, contributing to updating the dictionary adaptively. As the resulting cost function contains two nonsmooth (but proximable) terms, we approximate the latter regularizer by its Moreau envelope and apply the adaptive proximal forwardbackward splitting method to the approximated cost function. Numerical examples show the efficacy of the proposed algorithm.

AB - We propose an efficient multikernel adaptive filtering algorithm with double regularizers, providing a novel pathway towards online model selection and learning. The task is the challenging nonlinear adaptive filtering under no knowledge about a suitable kernel. Under this limited-knowledge assumption on an underlying model of a system of interest, many possible kernels are employed and one of the regularizers, a block ℓ1 norm for kernel groups, contributes to selecting a proper model (relevant kernels) in online and adaptive fashion, preventing a nonlinear filter from overfitting to noisy data. The other regularizer is the block ℓ1 norm for data groups, contributing to updating the dictionary adaptively. As the resulting cost function contains two nonsmooth (but proximable) terms, we approximate the latter regularizer by its Moreau envelope and apply the adaptive proximal forwardbackward splitting method to the approximated cost function. Numerical examples show the efficacy of the proposed algorithm.

KW - kernel adaptive filter

KW - multiple kernels

KW - proximity operator

UR - http://www.scopus.com/inward/record.url?scp=84901380295&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84901380295&partnerID=8YFLogxK

M3 - Conference contribution

SN - 9780992862602

BT - European Signal Processing Conference

PB - European Signal Processing Conference, EUSIPCO

ER -