Block-quantized kernel matrix for fast spectral embedding

The Internet environmenthas provided massive data to the actualindustrial production process. It not only has large amounts of databut also has a high data dimension, which brings challenges to thetraditional statistical process monitoring. Aiming at the nonlinearityand dynamics of industrial large-scale high-dimensional data, an efficientiterative multiple dynamic kernel principal component analysis (IMDKPCA)method is proposed to monitor the complex industrial process withsuper-large-scale high-dimensional data. In KPCA, a new KKT matrix is first created by using kernel matrix K. According to theproperties of the symmetric matrix, the newly constructed matrix hasthe same eigenvector as the original matrix K; hence, each columnof the matrix K can be used as the input sample of the iteration algorithm.After iterative operation, the kernel principal component can be deducedfleetly without the eigen decomposition. Because the kernel matrixis not stored in the algorithm beforehand, it can effectively reducethe computation complexity of the kernel. Especially for a tremendousdata scale, the traditional eigen decomposition technology is no longerappropriate, yet the presented method can be solved quickly. The autoregressivemoving average (ARMA) time series model and kernel principal componentanalysis (KPCA) are combined to build the IDKPCA model for dealingwith the dynamics and nonlinearity in the industrial process. Eventually,it is applied to monitor faults in the penicillin fermentation processand compared with MKPCA to certify the accuracy and applicabilityof the proposed method.

Kernel principal component analysis (KPCA) has become a popular technique for process monitoring in recent years. However, the performance largely depends on kernel function, yet methods to choose an appropriate kernel function among infinite ones have only been sporadically touched in the research literatures. In this paper, a novel methodology to learn a data-dependent kernel function automatically from specific input data is proposed and the improved kernel principal component analysis is obtained through using the data-dependent kernel function in traditional KPCA. The learning procedure includes two parts: learning a kernel matrix and approximating a kernel function. The kernel matrix is learned via a manifold learning method named maximum variance unfolding (MVU) which considers underlying manifold structure to ensure that principal components are linear in kernel space. Then, a kernel function is approximated via generalized Nystrom formula. The effectiveness of the improved KPCA model is confirmed by a numerical simulation and the Tennessee Eastman (TE) process benchmark.

In this paper, we introduce a method for multivariate function approximation using function evaluations, Chebyshev polynomials, and tensor-based compression techniques via the Tucker format. We develop novel randomized techniques to accomplish the tensor compression, provide a detailed analysis of the computational costs, provide insight into the error of the resulting approximations, and discuss the benefits of the proposed approaches. We also apply the tensor-based function approximation to develop low-rank matrix approximations to kernel matrices that describe pairwise interactions between two sets of points; the resulting low-rank approximations are efficient to compute and store (the complexity is linear in the number of points). We present an adaptive version of the function and kernel approximation that determines an approximation that satisfies a user-specified relative error over a set of random points. We extend our approach to the case where the kernel requires repeated evaluations for many values of (hyper)parameters that govern the kernel. We give detailed numerical experiments on example problems involving multivariate function approximation, low-rank matrix approximations of kernel matrices involving well-separated clusters of sources and target points, and a global low-rank approximation of kernel matrices with an application to Gaussian processes. We observe speedups up to 18X over standard matrix-based approaches.

Semi-supervised learning (SSL), as a powerful tool to learn from a limited number of labeled data and a large number of unlabeled data, has been attracting increasing attention in the machine learning community. In particular, the manifold regularization framework has laid solid theoretical foundations for a large family of SSL algorithms, such as Laplacian support vector machine (LapSVM) and Laplacian regularized least squares (LapRLS). However, most of these algorithms are limited to small scale problems due to the high computational cost of the matrix inversion operation involved in the optimization problem. In this paper, we propose a novel framework called Laplacian embedded regression by introducing an intermediate decision variable into the manifold regularization framework. By using ∈-insensitive loss, we obtain the Laplacian embedded support vector regression (LapESVR) algorithm, which inherits the sparse solution from SVR. Also, we derive Laplacian embedded RLS (LapERLS) corresponding to RLS under the proposed framework. Both LapESVR and LapERLS possess a simpler form of a transformed kernel, which is the summation of the original kernel and a graph kernel that captures the manifold structure. The benefits of the transformed kernel are two-fold: (1) we can deal with the original kernel matrix and the graph Laplacian matrix in the graph kernel separately and (2) if the graph Laplacian matrix is sparse, we only need to perform the inverse operation for a sparse matrix, which is much more efficient when compared with that for a dense one. Inspired by kernel principal component analysis, we further propose to project the introduced decision variable into a subspace spanned by a few eigenvectors of the graph Laplacian matrix in order to better reflect the data manifold, as well as accelerate the calculation of the graph kernel, allowing our methods to efficiently and effectively cope with large scale SSL problems. Extensive experiments on both toy and real world data sets show the effectiveness and scalability of the proposed framework.

It has been known in potential theory that, for some kernels matrices corresponding to well-separated point sets, fast analytical low-rank approximation can be achieved via the use of proxy points. This proxy point method gives a surprisingly convenient way of explicitly writing out approximate basis matrices for a kernel matrix. However, this elegant strategy is rarely known or used in the numerical linear algebra community. It still needs clear algebraic understanding of the theoretical background. Moreover, rigorous quantifications of the approximation errors and reliable criteria for the selection of the proxy points are still missing. In this work, we use contour integration to clearly justify the idea in terms of a class of important kernels. We further provide comprehensive accuracy analysis for the analytical compression and show how to choose nearly optimal proxy points. The analytical compression is then combined with fast rank-revealing factorizations to get compact low-rank approximations and also to select certain representative points. We provide the error bounds for the resulting overall low-rank approximation. This work thus gives a fast and reliable strategy for compressing those kernel matrices. Furthermore, it provides an intuitive way of understanding the proxy point method and bridges the gap between this useful analytical strategy and practical low-rank approximations. Some numerical examples help to further illustrate the ideas.

On using prototype reduction schemes to optimize kernel-based nonlinear subspace methods

Recently, the nonlinear projection trick (NPT) was introduced enabling direct computation of coordinates of samples in a reproducing kernel Hilbert space. With NPT, any machine learning algorithm can be extended to a kernel version without relying on the so called kernel trick. However, NPT is inherently difficult to be implemented incrementally because an ever increasing kernel matrix should be treated as additional training samples are introduced. In this paper, an incremental version of the NPT (INPT) is proposed based on the observation that the centerization step in NPT is unnecessary. Because the proposed INPT does not change the coordinates of the old data, the coordinates obtained by INPT can directly be used in any incremental methods to implement a kernel version of the incremental methods. The effectiveness of the INPT is shown by applying it to implement incremental versions of kernel methods such as, kernel singular value decomposition, kernel principal component analysis, and kernel discriminant analysis which are utilized for problems of kernel matrix reconstruction, letter classification, and face image retrieval, respectively.

This paper describes a novel approach for creating an efficient, general, and differentiable parameterization of large-scale non-Gaussian, non-stationary random fields (represented by multipoint geostatistics) that is capable of reproducing complex geological structures such as channels. Such parameterizations are appropriate for use with gradient-based algorithms applied to, for example, history-matching or uncertainty propagation. It is known that the standard Karhunen–Loeve (K–L) expansion, also called linear principal component analysis or PCA, can be used as a differentiable parameterization of input random fields defining the geological model. The standard K–L model is, however, limited in two respects. It requires an eigen-decomposition of the covariance matrix of the random field, which is prohibitively expensive for large models. In addition, it preserves only the two-point statistics of a random field, which is insufficient for reproducing complex structures. In this work, kernel PCA is applied to address the limitations associated with the standard K–L expansion. Although widely used in machine learning applications, it does not appear to have found any application for geological model parameterization. With kernel PCA, an eigen-decomposition of a small matrix called the kernel matrix is performed instead of the full covariance matrix. The method is much more efficient than the standard K–L procedure. Through use of higher order polynomial kernels, which implicitly define a high-dimensionality feature space, kernel PCA further enables the preservation of high-order statistics of the random field, instead of just two-point statistics as in the K–L method. The kernel PCA eigen-decomposition proceeds using a set of realizations created by geostatistical simulation (honoring two-point or multipoint statistics) rather than the analytical covariance function. We demonstrate that kernel PCA is capable of generating differentiable parameterizations that reproduce the essential features of complex geological structures represented by multipoint geostatistics. The kernel PCA representation is then applied to history match a water flooding problem. This example demonstrates that kernel PCA can be used with gradient-based history matching to provide models that match production history while maintaining multipoint geostatistics consistent with the underlying training image.

Principal component analysis (PCA) has been proven to be an efficient method in dimensionality reduction, feature extraction and pattern recognition. Kernel principal component analysis (KPCA) can be considered as a natural nonlinear generalization of PCA, which performs linear PCA in a high dimensional space implicitly by using kernel trick. However, both conventional PCA and KPCA suffer from the deficiency of being sensitive to outliers. Existing robust KPCA has to eigen-decompose the gram matrix directly in each step and is much more computationally infeasible due to the large size of the matrix when the number of training samples is large. By extending existing robust PCA algorithm using kernel methods, we present a novel robust adaptive algorithm for calculating the kernel principal components. The proposed method not only preserves the characteristic of capturing underlying nonlinear structure of KPCA but also is robust against outliers by restraining the effect of outlying samples. Compared with existing robust KPCA methods, our method is performed without having to store the kernel matrix, which can reduce significantly the storage burden. In addition, our method shows the potential of expansibility to the incremental learning version. Experimental results on synthetic data indicate that our improved algorithm is effective and promising.

Improved fault detection based on kernel PCA for monitoring industrial applications

Spectral clustering (SC) is a large family of grouping methods that partition data using eigenvectors of an affinity matrix derived from the data. Though SC methods have been successfully applied to a large number of challenging clustering scenarios, it is noteworthy that they will fail when there are significant intersections among different clusters. In this paper, based on the analysis that SC methods are able to work well when the affinity values of the points belonging to different clusters are relatively low, we propose a new method, called spectral multi-manifold clustering (SMMC), which is able to handle intersections. In our model, the data are assumed to lie on or close to multiple smooth low-dimensional manifolds, where some data manifolds are separated but some are intersecting. Then, local geometric information of the sampled data is incorporated to construct a suitable affinity matrix. Finally, spectral method is applied to this affinity matrix to group the data. Extensive experiments on synthetic as well as real datasets demonstrate the promising performance of SMMC.

The aim of this paper applies greedy kernel principal component analysis (greedy kernel PCA) to deal with training data reduction and nonlinear feature extraction in classification. Kernel PCA is a nonlinear extension of linear PCA. It shows a powerful nonlinear feature extraction technique via kernel trick. A disadvantage of kernel PCA, however, is that the storage of training data in terms of the dot products is too expensive since the size of kernel matrix increases quadratically with the number of training data. So, a more efficient feature extraction method, greedy kernel PCA, is proposed to reduce training data and nonlinear feature extraction for classification. The reduced set method aims to find a new kernel expansion and well approximates the original training data. Simulation results show both kernel PCA and greedy kernel PCA are more superior to linear PCA in feature extraction. Greedy kernel PCA will tend towards kernel PCA feature extraction as more percentage of training data is included in the reduced set, whilst greedy kernel PCA results in lower evaluation cost due to the reduced training set. The experiments show also that greedy kernel PCA can significantly reduce the complexity while retaining their accuracy in classification.

Machine Learning is a predominantly developing field in the current scenario due to spontaneous growth in size of data. Dimensionality Reduction is significant step used in machine learning. The Imprecation of dimensionality refers to the difficulty that comes up when working with high dimensional data. The Proposed work developed to reduce the risk of high dimensional data representation in the form of low dimensional data representation. This paper presents the dimensionality reduction practices such as Principal Component Analysis (PCA), Kernel PCA and Locally Linear Embedded (LLE).

We propose a new kernel entropy component analysis approach to object recognition with image set by fusing the information entropy. Since the geometry of Symmetric Positive Definite (SPD) matrices, we model the image set with its covariance matrix (nonsingular). Thus the object recognition with image set can be model as classifying problem on the riemannian manifold space. Given the proper kernel function derived from efficient metric for SPD matrices, points lie on manifold space represented by covariance matrices can be cast into high dimensional euclidean space. In this euclidean space, traditional methods used for dimensional reduction can be applied directly. Accounting for the measurability to information of the Renyi entropy, we fuse the information entropy to the dimensional reduction progress of our method. Similar to Kernel Principal Component Analysis (KPCA), the method accomplish data dimensionality reduction by projection onto a subset of entropy preserving KPCA axes. But this subset does not need to correspond to the top eigenvalues of the kernel matrix. The positive results of experiment on datasets demonstrated effectiveness of our method.

Data-driven techniques have been receiving considerable attention in the industrial process monitoring field due to their major advantages of easy implementation and less requirement for the prior knowledge and process mechanism. Principal component analysis (PCA) method is known as a popular method for monitoring and fault detection in industrial systems but as it is basically a linear method. However, most practical systems are nonlinear. To make the extension to nonlinear systems, kernel PCA (KPCA) method has been proposed for process modeling and monitoring. We present in this paper an online reduced rank optimized KPCA (RR-KPCA) technique for fault detection in order to extend the advantages of the KPCA models to online processes. Following the fault detection, the identification of the variables correlated to the fault occurred is of great importance. For this purpose, it is proposed to extend the approaches of localization by partial PCA and by elimination in the linear case to the nonlinear case, by exploiting the solution of reduction of the dimension of the kernel matrix in the feature space. The partial RR-KPCA and the elimination sensor identification (ESI-RRKPCA) are generated based on the static RR-KPCA and the online RR-KPCA methods. The idea of these approaches is to generate partial RR-KPCA models with reduced sets of variables. In other words, their goal is to generate indices of fault detection sensitive to certain faults and insensitive to others. The proposed fault isolation methods are applied for monitoring an air quality monitoring network (AIRLOR) data.