Fast instance selection for speeding up support vector machines

Chen, Jingnian; Zhang, Caiming; Xue, Xiaoping; Liu, Cheng‐Lin

doi:10.1016/j.knosys.2013.01.031

Cited by 50 publications

(27 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, not all the instances are useful for the training process, and there are also redundant instances that should be reduced. In fact, the instance selection is a common problem in the research of machine learning, or data mining, and many solutions have been proposed [32]. If some instances can be filtered out before the classifier training process, the efficiency of training process can be improved.…”

Section: The Instance Selection Problemsmentioning

confidence: 99%

A Hierarchical Method for Transient Stability Prediction of Power Systems Using the Confidence of a SVM-Based Ensemble Classifier

Zhou

et al. 2016

Energies

View full text Add to dashboard Cite

Machine learning techniques have been widely used in transient stability prediction of power systems. When using the post-fault dynamic responses, it is difficult to draw a definite conclusion about how long the duration of response data used should be in order to balance the accuracy and speed. Besides, previous studies have the problem of lacking consideration for the confidence level. To solve these problems, a hierarchical method for transient stability prediction based on the confidence of ensemble classifier using multiple support vector machines (SVMs) is proposed. Firstly, multiple datasets are generated by bootstrap sampling, then features are randomly picked up to compress the datasets. Secondly, the confidence indices are defined and multiple SVMs are built based on these generated datasets. By synthesizing the probabilistic outputs of multiple SVMs, the prediction results and confidence of the ensemble classifier will be obtained. Finally, different ensemble classifiers with different response times are built to construct different layers of the proposed hierarchical scheme. The simulation results show that the proposed hierarchical method can balance the accuracy and rapidity of the transient stability prediction. Moreover, the hierarchical method can reduce the misjudgments of unstable instances and cooperate with the time domain simulation to insure the security and stability of power systems.Energies 2016, 9, 778 2 of 20 from among massive sets of data, have been used to predict the transient stability [2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19]. The transient stability prediction can be treated as a two-class classification (stable and unstable) problem and solved by machine learning methods. A set of appropriate features/attributes is selected to generate the offline training sets, then an appropriate classification method is utilized to predict the transient stability status.There are two kinds of machine learning-based methods with different types of inputs [2]. One uses pre-fault steady-state variables as the original data, the machine learning method will be used to build the mapping between the steady-stable variables and the transient stability status with respect to an anticipated but not yet occurred contingency [2][3][4][5]. Once the current status is identified as insecure, preventive control can be carried out to modify the system to a secure state. However, when a serious fault happens, or failure of primary relay protection, the power system may still be unstable even if the preventive control has been conducted. Therefore we should emphasize importance of the study on transient stability prediction using post-fault responses. Because the post-fault responses carry information about the influence of the faults on the power system, the prediction is independent of the faults. This arises the second way of transient stability prediction based on machine learning techniques.With the development of wide-area measurement systems, the dynamic response of power systems...

show abstract

Section: The Instance Selection Problemsmentioning

confidence: 99%

A Hierarchical Method for Transient Stability Prediction of Power Systems Using the Confidence of a SVM-Based Ensemble Classifier

Zhou

et al. 2016

Energies

View full text Add to dashboard Cite

show abstract

“…Although other instance selecting algorithms can also reduce the size of original dataset ( e.g. , sparse modeling representative selection [18], multi-class instance selection [11]), they lack the mechanisms to train two biased classifiers. The second advantage of the proposed sample selection strategy is that confident classifiers trained on the two sets of points are able to predict confidently on opposite sides of the margin.…”

Section: Extensions Of Cpmmentioning

confidence: 99%

Confidence Preserving Machine for Facial Action Unit Detection

Zeng

Chu

Torre

et al. 2016

IEEE Trans. on Image Process.

View full text Add to dashboard Cite

Facial action unit (AU) detection from video has been a long-standing problem in automated facial expression analysis. While progress has been made, accurate detection of facial AUs remains challenging due to ubiquitous sources of errors, such as inter-personal variability, pose, and low-intensity AUs. In this paper, we refer to samples causing such errors as hard samples, and the remaining as easy samples. To address learning with the hard samples, we propose the Confidence Preserving Machine (CPM), a novel two-stage learning framework that combines multiple classifiers following an “easy-to-hard” strategy. During the training stage, CPM learns two confident classifiers. Each classifier focuses on separating easy samples of one class from all else, and thus preserves confidence on predicting each class. During the testing stage, the confident classifiers provide “virtual labels” for easy test samples. Given the virtual labels, we propose a quasi-semi-supervised (QSS) learning strategy to learn a person-specific (PS) classifier. The QSS strategy employs a spatio-temporal smoothness that encourages similar predictions for samples within a spatio-temporal neighborhood. In addition, to further improve detection performance, we introduce two CPM extensions: iCPM that iteratively augments training samples to train the confident classifiers, and kCPM that kernelizes the original CPM model to promote nonlinearity. Experiments on four spontaneous datasets GFT [15], BP4D [56], DISFA [42], and RU-FACS [3] illustrate the benefits of the proposed CPM models over baseline methods and state-of-the-art semisupervised learning and transfer learning methods.

show abstract

“…[12,9]), and (ii) Filter methods in which the selection criterion uses a selection function which is not based on a classifier (e.g., [29]). Most of the instance selection algorithms (e.g., [10,25,37,41]) are strongly related to the use of the k-NN classifier.…”

Section: Overviewmentioning

confidence: 99%