Narrow Results

This chapter examines the empirical performance of dynamic Gaussian affine term structure models (DGATSMs) at the zero lower bound (ZLB) when principal components analysis (PCA) is used to extract factors. We begin by providing a comprehensive review of DGATSM when PCA is used to extract factors highlighting its numerous auspicious qualities; it specifies bond yields to be a simple linear function of underlying Gaussian factors. This is especially favorable since, in principle, PCA works best when the model is linear and the first two moments are sufficient to describe the data, among other characteristics. DGATSM have a strong theoretical foundation grounded in the absence of arbitrage. DGATSM produce reasonable cross-sectional fits of the yield curve. Both of these qualities are inherited into the model when PCA is used to extract the state vector. Additionally, the implementation of PCA is simple in that it takes a matter of seconds to estimate factors and is convenient to include in estimation as most software packages have ready-to-use algorithms to compute the factors immediately. The results from our empirical investigation lead us to conclude that DGATSM, when PCA is employed to extract factors, perform very poorly at the ZLB. It frequently crosses the ZLB enroot to producing negative out-of-sample forecasts for bond yields. The main implication in this study is that despite its numerous positive characteristics, DGATSM when PCA is used to extract factors produce poor empirical forecasts around the ZLB.

Correlation Connected Clusters are objects that are grouped based upon correlations in local subsets of data. The method 'Computing Clusters of Correlation Connected objects' (4C)¹ uses DBSCAN and Principal Component Analysis (PCA) to find such objects. In this paper, we present a novel approach that attempts to find correlation connected clusters using an attribute partitioning approach to PCA. We prove that our novel approach is computationally superior to the 4C method.

In order to simplify the encoding process in vector quantization (VQ) image compression, a novel fast codebook search algorithm is presented in this paper. This algorithm integrates the technique of principle component analysis (PCA) with discrete wavelet transform (DWT) to quickly locate a reasonable initial codeword. Due to the smooth property of natural images, in addition, an adaptive search scope can be chosen by referring to the neighboring wavelet amplitudes and further used for the image block to be encoded. Experimental results showed that the proposed algorithm can indeed help providing a fast VQ encoder that only requires a few candidates for full comparison.

Producing good quality products is an important process control objective. However, achieving this objective can be very difficult in a continuous process, especially when quality measurements are not available on-line or they have long time delays. At the same time, process safety is also a critical issue. It is important to monitor process performance in real time. Here, a real time process quality monitoring and a control approach using multivariate statistical models are presented to achieve this objective. The goal of the monitoring approach is to detect faults in advance and the control approach is to decrease variations in product quality without real time quality measurements. A principal component analysis (PCA) type of model which incorporates time lagged variables is used, and the control objective is expressed in the score space of this PCA model. A process quality monitor is developed based on the process description of this PCA model. A controller is designed in the model predictive control (MPC) framework, and it is used to control the equivalent score space representation of the process. The score predictive score model for the MPC algorithm is built using partial least squares (PLS). The proposed controller can be developed from and implemented on top of existing PID control systems. The proposed monitor and controller are demonstrated in case studies, which involve a binary distillation column and the Tennessee Eastman process.

As one of the important research directions of Intelligent Transportation System (ITS), Traffic Sign Recognition (TSR) has become a hot topic worldwide. Due to the wide variety of traffic signs and complexity of road traffic environment, efficiency and correct recognition rate have to be taken into consideration in the design of TSR schemes. This paper uses principal component analysis (PCA) and extreme learning machine (ELM) to recognize traffic signs. Firstly histograms of the oriented gradient (HOG) features of each traffic sign are extracted, after dimension reduction by PCA, the dimension-reduced PCA features are put into ELM to train for an optimized recognition model. Database used in the experiments is German Traffic Sign Detection Benchmark (GTSDB). Experimental results show that the proposed method is able to recognize traffic sign in real-time with a high correct recognition rate.

In order to improve the efficiency and accuracy of transformer fault diagnosis, a new fault diagnosis method based on PCA and BP neural network is proposed. The proposed method used PCA to reduce the data of fault messages dimensions through matrix conversion, and then used BP neural network for transform fault diagnosis. Illustrated by examples, in the fault diagnosis of transformer, the PCA can concentrate fault information without influencing analysis results under the condition that the loss of useful messages is relatively small. Based on that, compared with RBF and GRNN neural network, the BP neural network has a higher accuracy for transform fault diagnosis. As a consequence, combination of PCA and BP neural network is a practical and feasible method in transform fault diagnosis.

Head detection plays an important role in pedestrian detection for the unique feature of human head. The existing methods of head detection mainly use the features of contour, color and template, which usually have low accuracy and robustness of detection. In this paper, a novel method based on CNN with multi-stage feature integration to detect head in complex background is proposed. This method combines the shallow and deep feature of CNN which can increase the completeness and comprehensiveness of image description. Meanwhile, the computation cost is decreased by reduced high dimensional features with PCA. The experimental results show that the proposed method has high detection accuracy, which is better than the existing methods.

Consensus reaching processes (CRPs) in Group Decision Making (GDM) try to reach a mutual agreement among a group of decision makers before making a decision. To evaluate and understand the performance of a CRP is often complex due to, mainly, the presence of disagreement among decision makers. A clear, simple, correct and suitable visualization of the discussion consensus rounds is key for facilitating the analysis of such performance because, without a clear visualization, it is hard to understand the disagreements among experts. This paper proposes a new visualization related to experts’ preferences and their evolution for CRPs based on the Principal Component Analysis (PCA).

A classification method of discriminate rice from different varieties with voltammetric electronic tongue based on square wave voltammetry is investigated. The rice samples are crushed and mixed with distilled water to get the rice solution, and the solution should be stirred and filtered before the experiment. In order to obtain the electrochemical response signals of the rice samples and extract the characteristic value of the singles, the electronic tongue which works respectively with titanium (Ti) electrode and tungsten electrode (W) to test the sample solution under square wave voltammetry. The Principal Component Analysis (PCA) and Clustering Analysis (CA) are adopted to classify and recognize the rice samples. Experimental results show that good classification and recognition results are got in this paper when using Principal Component Analysis and Cluster Analysis to analyze the response signals which are obtained by voltammetric electronic tongue worked with Ti electrode and W electrode under square wave potential.

This chapter examines the empirical performance of dynamic Gaussian affine term structure models (DGATSMs) at the zero lower bound (ZLB) when principal components analysis (PCA) is used to extract factors. We begin by providing a comprehensive review of DGATSM when PCA is used to extract factors highlighting its numerous auspicious qualities; it specifies bond yields to be a simple linear function of underlying Gaussian factors. This is especially favorable since, in principle, PCA works best when the model is linear and the first two moments are sufficient to describe the data, among other characteristics. DGATSM have a strong theoretical foundation grounded in the absence of arbitrage. DGATSM produce reasonable cross-sectional fits of the yield curve. Both of these qualities are inherited into the model when PCA is used to extract the state vector. Additionally, the implementation of PCA is simple in that it takes a matter of seconds to estimate factors and is convenient to include in estimation as most software packages have ready-to-use algorithms to compute the factors immediately. The results from our empirical investigation lead us to conclude that DGATSM, when PCA is employed to extract factors, perform very poorly at the ZLB. It frequently crosses the ZLB enroot to producing negative out-of-sample forecasts for bond yields. The main implication in this study is that despite its numerous positive characteristics, DGATSM when PCA is used to extract factors produce poor empirical forecasts around the ZLB.