Papers by Keyword: Dimension Reduction

Abstract: In face recognition algorithms, Principal Component Analysis (PCA) is one of classical algorithms. But PCA algorithm needs to convert each sample matrix into vectors, which leads to a large amount of calculations in solving high-rank matrix. The essence of Modular Two-dimensional Principle Component Analysis (2DPCA) is that original images are divided into modular images, and image covariance matrix is constructed directly from the sub-images by the optimal projection matrix. But the number of features is still large and correlation still exists in feature extraction, which influences the speed of classification. In order to solve this problem, we proposed a method combining the Modular 2DPCA with PCA to reduce the dimension of features and decrease the correlation among feature parameters. The experimental results based on ORL Human Face Database show that the recognition rate of the algorithm is superior to single Modular 2DPCA or PCA.

Abstract: Feature selection is an effective pre-processing technology to facilitate text mining on high dimensional feature space. In recent years, many effective redundant feature selection methods have been proposed from different motivations. However, a comparative experimental study on redundant feature selection methods in the field of text mining has not been reported yet. In order to solve this problem, an extensive empirical comparative study with the task of text classification is given in the paper. The experimental results indicate that the 3-way Mutual Information represents the redundancy much better than traditional 2-way Mutual Information, since the label information are considered by 3-way Mutual Information. As a result, the performances of redundant feature selection methods based on 3-way Mutual Information overwhelm other methods.

Abstract: Assembly processes are often complex and highly nonlinear. In sheet metal assemblies, the most important factor that makes the process nonlinear is the contact interaction between mating parts during the assembly process. This paper focuses on developing a methodology for nonlinear variation analysis of compliant sheet metal assemblies which also includes the effect of surface continuity of components. The proposed methodology integrates a nonlinear finite element analysis with an improved sensitivity-free probability analysis in order to predict the final assembly variation. The efficiency of the developed approach is evaluated by an experimental case study as well as Monte Carlo simulation.

Abstract: LLE is a very effective non-linear dimension reduction algorithm and widely explored in machine learning, pattern recognition, data mining and etc. Locally linear, Globally non-linear has always been regarded as the features and advantages of LLE. However, the theoretical derivation presented in this paper shows that when the size of neighborhood is larger than the dimension of the space in which the data are presented, LLE is no longer global nonlinear and almost has the same effect as PCA in dimensionality reduction. At present, a lot of literatures on LLE verify their results on Swiss Roll, Punctured Sphere, Twin Peaks, etc. These manifolds are presented in the three-dimensional Euclidean space and the size of neighborhood is always larger than three to prevent too small to be effective. But in these cases, LLE cannot play its advantage of nonlinearity.

Abstract: In order to short the modelling time of BP neural network, this paper designs a kind of genetic algorithm to optimize it. By encoding the individual components, initializing the number of populations, and designing proper fitness function, a binary coding genetic algorithm is provided for BP neural network. And it is used to optimize input variables of BP neural network and reduce its dimension. The experiment is carried out based on KDD Cup 99 data set. The results show that the optimized model has shorter modelling time.

Abstract: In order to improve the diagnosis performance of the types of Loss of Coolant Accident (LOCA), A diagnose model based on manifold Learning is built. As the manifold learning method using to reduce the observed parameters of reactors dimensions, the ANN is used to get dimension reducing mapping and modes classifying method. This method improves the ability of the model to identify as well as the robustness of it. The experiment results show that the systems diagnosis precision is high and the key parameters are analyzed efficiently.

Abstract: The intelligent diagnosis emphasizes the processing method of knowledge of the historical data. The capability of an intelligent diagnosis system depends on the knowledge possessed by the system, especially by the specific knowledge in application. Currently, most of the important equipment have their own the inspection systems. With the help of these systems, plenty of historical data can be collected in real time. This paper discusses the possibility of the application of similarity extraction and pattern discovery of time series in fault diagnosis by using these historical data, presents the method of time series feature extraction and pattern matching, and advances the possibility of data clustering and pattern discovery based on dimension reduction.

Abstract: This paper describes a dimension reduction method of input vector to improve classification efficiency of LVQ neural network, where GA is used to decrease the redundancy of input data. And in order to solve the initial weight vector sensitivity, GA is also employed to optimize the initial vector. The experimental results on the UCI data sets demonstrate that the efficiency and accuracy of our LVQ network by GA is higher than general LVQ neural network classification algorithm.

Abstract: Kernel entropy component analysis (KECA) reveals the original data’s structure by kernel matrix. This structure is related to the Renyi entropy of the data. KECA maintains the invariance of the original data’s structure by keeping the data’s Renyi entropy unchanged. This paper described the original data by several components on the purpose of dimension reduction. Then the KECA was applied in celestial spectra reduction and was compared with Principal Component Analysis (PCA) and Kernel Principal Component Analysis (KPCA) by experiments. Experimental results show that the KECA is a good method in high-dimensional data reduction.

Abstract: Nonnegative Matrix Factorization (NMF) is among the most popular subspace methods, widely used in a variety of image processing problems. However, this approach is very time-consuming in face recognition due to the extreme high dimensionality of the original matrix. To remedy this limitation, this paper presents a Decorrelation-based NMF (DNMF) method. The proposed algorithm first takes into account the dimension reduction of the original matrix by preprocessing of decorrelation in spatial domain, and then uses nearest neighbor classifier on the reduced subspace. The developed algorithm has been applied for the ORL standard face image database. Experimental results demonstrate the validity of this method.