文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Nonlinear Regularization with Prior Templates
侵权投诉

Nonlinear Regularization with Prior Templates

A general method to include prior knowledge in regularization functionals is presented. In a first step, components of prior information are formalized by defining relevant distances from template functions, playing the role of locally or

Nonlinear Regularization with Prior Templates Jorg C. Lemm Institut fur Theoretische Physik I Wilhelm{Klemm{Str.9 D{48149 Munster Germany Email: lemm@uni-muenster.de

Extended Abstract

1 Introduction A general method to include prior knowledge in regularization functionals is presented. In a rst step, components of prior information are formalized by de ning relevant distances from template functions, playing the role of locally or globally de ned function prototypes. In a second step the components are combined by the logical operations AND and OR, similar to fuzzy logical methods. Implementing the probabilistic AND yields the classical linear regularization methods, like splines or Radial Basis Functions. For many problems knowledge formulated in terms of a probabilistic OR is available. Objects, like faces, for example consist of constituents, e.g. eyes, mouth, nose, which may appear in one of several relatively well de ned variants, e.g. eyes may be open OR closed, blue OR brown. The implementation of such a probabilistic OR, however, yields nonlinear stationarity equations, which are the topic of the present paper. Two approaches are introduced for implementing a probabilistic OR. The rst one interprets the individual components as (shifted and rescaled) log{ probabilities and results in a mixture model over a function space, resembling a physical system at nite temperature. The corresponding temperature can be varied, similarly to the approach of Rose, Gurewitz,& Fox (1990) for clustering. The second approach directly constructs an e ective total posterior by fuzzy logical{like combination of prior components. Introducing a temperature{like parameter this corresponds to the Landau{Ginzburg treatment of phase transitions. The resulting equations turn out to be similar, for example, to equations obtained in quantum mechanical scattering theory. For both methods a variety of local learning algorithms can be chosen. The two approaches and corresponding learning algorithms are exempli ed in a numerical model. 1

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email