ISSN:0886-9383    

DOI:10.1002/cem.1180080603

出版商:John Wiley&Sons, Ltd.    

年代:1994

数据来源: WILEY

摘要:

AbstractMultivariate images are very large data structures and any type of regression for their analysis is very computer‐intensive. Kernel‐based partial least squares (PLS) regression, presented in an earlier paper, makes the calculation phase more rapid and less demanding in computer memory. The present paper is a direct continuation of the first paper. In this study the kernel PLS algorithm is extended to include cross‐validation for determination of the optimal model dimensionality. To show the applicability of the kernel algorithm, two examples from multivariate image analysis are used. The first example is an image from an airborne scanner of size 9 × 512 × 512. It consists of nine images which are regressed against a constructed dependent image to test the accuracy of the kernel algorithm when used on large data structures. The second example is a satellite image of size 7 × 512 × 512. Several different regression models are presented together with a comparison of their predictive capabilities. The regression models are also used as examples for showing the use of cross

ISSN:0886-9383    

DOI:10.1002/cem.1180080604

出版商:John Wiley&Sons, Ltd.    

年代:1994

数据来源: WILEY

摘要:

AbstractFinding methods for the optimization of weights in feedforward neural networks has become an ongoing developmental process in connectionist research. The current focus on finding new methods for the optimization of weights is mostly the result of the slow and unreliable convergence properties of the gradient descent optimization used in the original back‐propagation algorithm. More accurate and computationally expensive second‐order gradient methods have displaced earlier first‐order gradient optimization of the network connection weights. The global, extended Kalman filter is among the most accurate and computationally expensive of these second‐order weight optimization methods. The iterative, second‐order nature of the filter results in a large number of calculations for each sweep of the training set. This can increase the training time dramatically when training is conducted with data sets that contain large numbers of training patterns. In this paper an adaptive variant of the global, extended Kalman filter that exhibits substantially improved convergence properties is presented and discussed. The adaptive mechanism permits more rapid convergence of network training by identifying data that contain redundant information and avoiding calculations based on this redundant in

ISSN:0886-9383    

DOI:10.1002/cem.1180080605

出版商:John Wiley&Sons, Ltd.    

年代:1994

数据来源: WILEY

摘要:

AbstractIn a variety of spectroscopic techniques the fundamental problem exists of determination of the individual spectral components, intrinsically overlapped in the measured spectrum. This is a typical deconvolution problem and several methods and techniques have been proposed for its solution in the technical literature, but suggestions of new approaches are still of interest. A new deconvolution procedure is presented here based on the use of the conjugate gradient minimization algorithm with the addition of sutiable constraints directly obtained by the application to the measured spectrum of the second‐derivative operator or more sophisticated resolution enhancement procedures. Since in the examined case deconvolution essentially requires the minimization of a non‐convex function, the use of such constraints is extremely important to supply suitable input parameters to the conjugate gradient algorithm to avoid obtaining minimum points which have no physical meaning. In our case each spectral compoent used for deconvolution has been assumed to have a Gaussian analytical definition fully identified by three parameters (amplitude, central frequency, spectral bandwidth), so that the input values required to start the deconvolution process are the numberMof Gaussian components and 3Msuitable initial approximations of the parameters above. It is shown that all this information can be obtained from the measured data. The deconvolution procedure was implemented by a FORTRAN Microsoft Version 5.1 program and experimental results relative to spectroscopic data obtained by FT‐IR analysis of human serum albumin are reported. The results are discussed and compared with data obtained by the use of other techn

ISSN:0886-9383    

DOI:10.1002/cem.1180080606

出版商:John Wiley&Sons, Ltd.    

年代:1994

数据来源: WILEY

摘要:

AbstractA method for selecting variables in the non‐sequential direction of a two‐way data matrix (e.g. wavelength in diode array HPLC) is described. Composition 1 and 2 resolvability indices are calculated according to the size of eigenvalues of uncentred data matrices as a window is moved along the sequential direction. A double‐window technique is then performed where resolvability indices are calculated as a window is moved along the non‐sequential direction. Some regions have higher resolvability indices and hence are more useful for resolution. Variables are ranked according to resolvability. Two simulations are analysed and it is shown that it is possible to obtain good resolution on a small subset of the original va

ISSN:0886-9383    

DOI:10.1002/cem.1180080607

出版商:John Wiley&Sons, Ltd.    

年代:1994

数据来源: WILEY

摘要:

AbstractAn inverse mapping method called PCB (principal component backing), in which the point representing an unknown sample from a low‐dimensional principal component subspace is back‐projected to the high‐dimensional original feature space, is proposed. Two sorts of boundary conditions, non‐linear inverse mapping and linear inverse mapping, are used to obtain an accurate solution in the PCB method. The method is applied to the material design of high‐Tcsuperconductors, predicting the composition and process conditions for the synthesis of F‐doped Bi‐based materials. Samples in the ‘optimal’ region with the highestTcof the Bi‐based ceramics

ISSN:0886-9383    

DOI:10.1002/cem.1180080608

出版商:John Wiley&Sons, Ltd.    

年代:1994

数据来源: WILEY

ISSN:0886-9383    

DOI:10.1002/cem.1180080609

出版商:John Wiley&Sons, Ltd.    

年代:1994

数据来源: WILEY

ISSN:0886-9383    

DOI:10.1002/cem.1180080610

出版商:John Wiley&Sons, Ltd.    

年代:1994

数据来源: WILEY

ISSN:0886-9383    

DOI:10.1002/cem.1180080611

出版商:John Wiley&Sons, Ltd.    

年代:1994

数据来源: WILEY