this paper we study Tikhonov regularization (Tikhonov and Arsenin 1977). While Tikhonov regularization has had significant impact on several branches of science and engineering dealing with ill-posed and inverse problems, in particular modeling and analysis of high-dimensional or distributed signals and data (Tikhonov and Arsenin 1977, O'Sullivan 1986, Wahba 1990, Poggio

: Prior knowledge can be introduced into system identification problems in terms of constraints on the parameter space, or regularizing penalty functions in a prediction error criterion. The contribution of this work is mainly an extention of the well known FPE (Final Prediction Error) statistic to the case when the system identification problem is constrainted and contains a regularization penalty. The FPECR statistic (Final Prediction Error with Constraints and Regularization) is of potential interest as a criterion for selection of both regularization parameters and structural parameters such as order. Keywords: Regularization, Optimization, Parameter Estimation, Nonlinear Systems. 1. INTRODUCTION In practical system identification it is often desirable to introduce prior knowledge into the problem, rather than relying completely on the data. If the model structure is assumed to be fixed, there are still several approaches, cf. Figure 1: (1) Constraints on the parameter space, for e...

As the Environmental Protection Agency has indicated in Emission Inventory Improvement Program (EIIP) documents, the choice of methods to be used to estimate emissions depends on how the estimates will be used and the degree of accuracy required. Methods using site-specific data are preferred over other methods. These documents are non-binding guidance and not rules. EPA, the States, and others retain the discretion to employ or to require other approaches that meet the requirements of the applicable statutory or regulatory requirements in individual

Abstract not found

Time series are a class of data whose complexity and rich structure make it difficult for data mining tools to extract meaningful patterns from them, and in particular to prune away the false positive patterns. Wavelet-based methods have recently become the preferred way for significance testing of time series and time series collections, but these methods are still often based on fairly ad hoc bootstrapping techniques in the wavelet domain without a disciplined null model analysis. We propose a new well-grounded null model for time series collections that also sets minimum requirements for realistic resampling methods. We compare it to the null models of common resampling methods and introduce a new randomization method that is compatible with the proposed null model. We conduct experiments on real and synthetic datasets to compare the behavior of the various methods and reflect the results to the differences in their null models. Compared with the other methods, our experiments suggest that the proposed method gives fewer Type I and Type II errors across a range of statistics.