原创力文档- 1、本文档共5页，可阅读全部内容。
- 2、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。
- 3、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击 这里二次下载。
- 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。
查看更多
A kernel-based RLS algorithm for nonlinear adaptive filtering using
A kernel-based RLS algorithm for nonlinear adaptive filtering
using sparse approximation theory
Ce?dric Richard
Institut Charles Delaunay (ICD, FRE CNRS 2848), Laboratoire LM2S
Universite? de Technologie de Troyes, BP 2060, 10010 Troyes cedex - France
tel.: +33.3.25.71.58.47 fax.: +33.3.25.71.56.99 cedric.richard@utt.fr
1 Short abstract
In the last ten years, there has been an explosion of activity in the field of learning algorithms utilizing
reproducing kernels, most notably for classification and regression. A common characteristic in kernel-
based methods is that they deal with models whose order equals the number of input data, making them
unsuitable for online applications. In this paper, we investigate a new kernel-based RLS algorithm that
makes unnecessary the use of any computationally demanding sparsification procedure. The increase
in the model order is controlled by the coherence parameter, a fundamental quantity that is used to
characterize the behavior of dictionaries in sparse approximation problems.
2 Extended abstract
Adaptive filtering has become a topic of keen interest over the past three decades to help cope with
time variations of system parameters and lack of a priori statistical information [11, 15]. Linear models
are still routinely used because of their inherent simplicity from conceptual and implementational point
of view. In many practical situations, however, nonlinear signal processing is needed. It includes items
such as nonlinear system identification, prediction and control, e.g., in communications and biomedical
engineering, see [9]. Following the pioneering works [1, 2, 13], there has been recent progress in function
approximation methods based on reproducing kernel Hilbert spaces (RKHS) [12, 16], including, for
example, support vector regression [18]. A common characteristic in kernel-based methods is that
they deal with models whose order is the size of the training set, making them unsuitable for online
applications. Several alg
文档评论（0）