Adaptive dimensionality reduction for neural network-based online principal component analysis

“Principal Component Analysis” (PCA) is an established linear technique for dimensionality reduction. It performs an orthonormal transformation to replace possibly correlated variables with a smaller set of linearly independent variables, the so-called principal components, which capture a large portion of the data variance. The problem of finding the optimal number of principal components has been widely studied for offline PCA. However, when working with streaming data, the optimal number changes continuously. This requires to update both the principal components and the dimensionality in every timestep. While the continuous update of the principal components is widely studied, the available algorithms for dimensionality adjustment are limited to an increment of one in neural network-based and incremental PCA. Therefore, existing approaches cannot account for abrupt changes in the presented data. The contribution of this work is to enable in neural network-based PCA the continuous dimensionality adjustment by an arbitrary number without the necessity to learn all principal components. A novel algorithm is presented that utilizes several PCA characteristics to adaptivly update the optimal number of principal components for neural network-based PCA. A precise estimation of the required dimensionality reduces the computational effort while ensuring that the desired amount of variance is kept. The computational complexity of the proposed algorithm is investigated and it is benchmarked in an experimental study against other neural network-based and incremental PCA approaches where it produces highly competitive results.

A subspace type incremental two-dimensional principal component analysis algorithm

Principal component analysis (PCA) has been a powerful tool for high-dimensional data analysis. It is usually redesigned to the incremental PCA algorithm for processing streaming data. In this paper, we propose a subspace type incremental two-dimensional PCA algorithm (SI2DPCA) derived from an incremental updating of the eigenspace to compute several principal eigenvectors at the same time for the online feature extraction. The algorithm overcomes the problem that the approximate eigenvectors extracted from the traditional incremental two-dimensional PCA algorithm (I2DPCA) are not mutually orthogonal, and it presents more efficiently. In numerical experiments, we compare the proposed SI2DPCA with the traditional I2DPCA in terms of the accuracy of computed approximations, orthogonality errors, and execution time based on widely used datasets, such as FERET, Yale, ORL, and so on, to confirm the superiority of SI2DPCA.

Feature Snatching and Performance Assessment for Connoting the Admittance Likelihood of student using Principal Component Reduction

Recently, engineers are concentrating on designing an effective prediction model for finding the rate of student admission in order to raise the educational growth of the nation. The method to predict the student admission towards the higher education is a challenging task for any educational organization. There is a high visibility of crisis towards admission in the higher education. The admission rate of the student is the major risk to the educational society in the world. The student admission greatly affects the economic, social, academic, profit and cultural growth of the nation. The student admission rate also depends on the admission procedures and policies of the educational institutions. The chance of student admission also depends on the feedback given by all the stake holders of the educational sectors. The forecasting of the student admission is a major task for any educational institution to protect the profit and wealth of the organization. This paper attempts to analyze the performance of the student admission prediction by using machine learning dimensionality reduction algorithms. The Admission Predict dataset from Kaggle machine learning Repository is used for prediction analysis and the features are reduced by feature reduction methods. The prediction of the chance of Admit is achieved in four ways. Firstly, the correlation between each of the dataset attributes are found and depicted as a histogram. Secondly, the top most high correlated features are identified which are directly contributing to the prediction of chance of admit. Thirdly, the Admission Predict dataset is subjected to dimensionality reduction methods like principal component analysis (PCA), Sparse PCA, Incremental PCA , Kernel PCA and Mini Batch Sparse PCA. Fourth, the optimized dimensionality reduced dataset is then executed to analyze and compare the mean squared error, Mean Absolute Error and R2 Score of each method. The implementation is done by python in Anaconda Spyder Navigator Integrated Development Environment. Experimental Result shows that the CGPA, GRE Score and TOEFL Score are highly correlated features in predicting the chance of admit. The execution of performance analysis shows that Incremental PCA have achieved the effective prediction of chance of admit with minimum MSE of 0.09, MAE of 0.24 and reasonable R2 Score of 0.26.

Object Tracking Based on Online Semi-Supervised SVM and Adaptive-Fused Feature

In order to improve the performance of tracking, we propose a new online tracking method based on classification and adaptive fused feature. We first label a few positive and negative samples, train the classifier by the online SSSM (Semi-Supervised Support Vector Machine) learning and these labelled samples, and then locate the position of the object from the next frame according to the trained classifier. In order to adapt more of the new samples, we need to update the classifier by finding new samples with high confident value obtained by the trained classifier and add them into the online SSSM. Finally we also update the object model by the online incremental PCA (Principal Component Analysis) because of background clutter, heavy occlusion and complicated object appearance changes. Compared with the basic mean shift tracking and the ensemble tracking method, experimental results show that our tracking method is able to effectively handle heavy occlusion and background clutter in some challenge videos including some thermal videos.