Results 1 - 10
of
13
A Variance Minimization Criterion to Active Learning on Graphs
"... We consider the problem of active learning over the vertices in a graph, without feature representation. Our study is based on the common graph smoothness assumption, which is formulated in a Gaussian random field model. We analyze the probability distribution over the unlabeled vertices conditioned ..."
Abstract
-
Cited by 17 (3 self)
- Add to MetaCart
(Show Context)
We consider the problem of active learning over the vertices in a graph, without feature representation. Our study is based on the common graph smoothness assumption, which is formulated in a Gaussian random field model. We analyze the probability distribution over the unlabeled vertices conditioned on the label information, which is a multivariate normal with the mean being the harmonic solution over the field. Then we select the nodes to label such that the total variance of the distribution on the unlabeled data, as well as the expected prediction error, is minimized. In this way, the classifier we obtain is theoretically more robust. Compared with existing methods, our algorithm has the advantage of selecting data in a batch offline mode with solid theoretical support. We show improved performance over existing label selection criteria on several real world data sets. 1
Online Feature Selection with Streaming Features
, 2012
"... We propose a new online feature selection framework for applications with streaming features where the knowledge of the full feature space is unknown in advance. We define streaming features as features that flow in one by one over time whereas the number of training examples remains fixed. This is ..."
Abstract
-
Cited by 8 (4 self)
- Add to MetaCart
We propose a new online feature selection framework for applications with streaming features where the knowledge of the full feature space is unknown in advance. We define streaming features as features that flow in one by one over time whereas the number of training examples remains fixed. This is in contrast with traditional online learning methods that only deal with sequentially added observations, with little attention being paid to streaming features. The critical challenges for online streaming feature selection include (1) the continuous growth of feature volumes over time; (2) a large feature space, possibly of unknown or infinite size; and (3) the unavailability of the entire feature set before learning starts. In the paper, we present a novel Online Streaming Feature Selection (OSFS) method to select strongly relevant and nonredundant features on the fly. An efficient Fast-OSFS algorithm is proposed to improve feature selection performance. The proposed algorithms are evaluated extensively on high-dimensional datasets and also with a real-world case study on impact crater detection. Experimental results demonstrate that the algorithms achieve better compactness and higher prediction accuracy than existing streaming feature selection algorithms.
Embedding Motion and Structure Features for Action Recognition
, 2013
"... We propose a novel method to model human actions by explicitly coding motion and structure features that are separately extracted from video sequences. Firstly, the motion template (one feature map) is applied to encode the motion information and image planes (five feature maps) are extracted from ..."
Abstract
-
Cited by 4 (3 self)
- Add to MetaCart
(Show Context)
We propose a novel method to model human actions by explicitly coding motion and structure features that are separately extracted from video sequences. Firstly, the motion template (one feature map) is applied to encode the motion information and image planes (five feature maps) are extracted from the volume of differences of frames to capture the structure information. The Gaussian pyramid and center-surround operations are performed on each of the six obtained feature maps, decomposing each feature map into a set of subband maps. Biologically inspired features are then extracted by successively applying Gabor filtering and max pooling on each subband map. To make a compact representation, discriminative locality alignment is employed to embed the high-dimensional features into a low-dimensional manifold space. In contrast to sparse representations based on detected interest points, which suffer from the loss of structure information, the proposed model takes into account the motion and structure information simultaneously and integrates them in a unified framework; it therefore provides an informative and compact representation of human actions. The proposed method is evaluated on the KTH, the multiview IXMAS, and the challenging UCF sports datasets and outperforms state-of-the-art techniques on action recognition.
Improving Semi-Supervised Target Alignment via Label-Aware Base Kernels
"... Semi-supervised kernel design is an essential step for obtain-ing good predictive performance in semi-supervised learning tasks. In the current literatures, a large family of algorithms builds the new kernel by using the weighted average of pre-defined base kernels. While optimal weighting schemes h ..."
Abstract
-
Cited by 1 (1 self)
- Add to MetaCart
(Show Context)
Semi-supervised kernel design is an essential step for obtain-ing good predictive performance in semi-supervised learning tasks. In the current literatures, a large family of algorithms builds the new kernel by using the weighted average of pre-defined base kernels. While optimal weighting schemes have been studied extensively, the choice of base kernels received much less attention. Many methods simply adopt the empiri-cal kernel matrices or its eigenvectors. Such base kernels are computed irrespective of class labels and may not always re-flect useful structures in the data. As a result, in case of poor base kernels, the generalization performance can be degraded however hard their weights are tuned. In this paper, we pro-pose to construct high-quality base kernels with the help of label information to globally improve the final target align-ment. In particular, we devise label-aware kernel eigenvec-tors under the framework of semi-supervised eigenfunction extrapolation, which span base kernels that are more useful for learning. Such base kernels are individually better aligned to the learning target, so their mixture will more likely gen-erate a good classifier. Our approach is computationally ef-ficient, and demonstrates encouraging performance in semi-supervised classification and regression.
Feature Extraction and Representation for Human Action Recognition
, 2013
"... Human action recognition, as one of the most important topics in computer vision, has been extensively researched during the last decades; however, it is still regarded as a challenging task especially in realistic scenarios. The difficulties mainly result from the huge intra-class variation, backgr ..."
Abstract
- Add to MetaCart
Human action recognition, as one of the most important topics in computer vision, has been extensively researched during the last decades; however, it is still regarded as a challenging task especially in realistic scenarios. The difficulties mainly result from the huge intra-class variation, background clutter, occlusions, illumination changes and noise. In this thesis, we aim to enhance human action recognition by feature extraction and representation using both holistic and local methods. Specifically, we have first proposed three approaches for the holistic representa-tion of actions. In the first approach, we explicitly extract the motion and structure features from video sequences by converting the video representation into a 2D im-age representation problem; In the second and third approaches, we treat the video sequences as 3D volumes and propose to use spatio-temporal pyramid structures to extract multi-scale global features. Gabor filters and steerable filters are extended to the video domain for holistic representations, which have been demonstrated to be successful for action recognition. With regards to local representations, we have first-
Feature Selection using Eigenvalue Optimization and Partial Least Squares
"... Feature selection is an essential problem in many fields such as computer vision. In this paper we introduce a supervised feature selection criterion based on Partial Least Squares regression (PLS). We find an optimal feature subset by applying the theory of Optimal Experiment Design to optimize the ..."
Abstract
- Add to MetaCart
(Show Context)
Feature selection is an essential problem in many fields such as computer vision. In this paper we introduce a supervised feature selection criterion based on Partial Least Squares regression (PLS). We find an optimal feature subset by applying the theory of Optimal Experiment Design to optimize the eigenvalues of the loadings matrix obtained from PLS. Since PLS extracts components such that the covariance between features and response variable and the covariance between features itself are simultaneously maximized, the criterion simultaneously satisfies the relevance property towards the response variable and the latent information in features. In order to optimize the eigenvalues, we use the D-optimality criterion which maximizes the determinant of loadings covariance matrix. The paper first introduces a theoretical proof of the proposed criterion followed by empirical evaluation using two image data sets. Our experiments demonstrate that our Optimal Loadings criterion outperforms other popular supervised feature selection techniques. 1.
Local and Global Discriminative Learning for Unsupervised Feature Selection
"... Abstract—In this paper, we consider the problem of feature selection in unsupervised learning scenario. Recently, spectral feature selection methods, which leverage both the graph Laplacian and the learning mechanism, have received consid-erable attention. However, when there are lots of irrelevant ..."
Abstract
- Add to MetaCart
(Show Context)
Abstract—In this paper, we consider the problem of feature selection in unsupervised learning scenario. Recently, spectral feature selection methods, which leverage both the graph Laplacian and the learning mechanism, have received consid-erable attention. However, when there are lots of irrelevant or noisy features, such graphs may not be reliable and then mislead the selection of features. In this paper, we propose the Local and Global Discriminative learning for unsupervised Feature Selection (LGDFS), which integrates a global and a set of locally linear regression model with weighted 2-norm regularization into a unified learning framework. By exploring the discriminative and geometrical information in the weighted feature space, which alleviates the effects of the irrelevant features, our approach can find the most representative features to well respect the cluster structure of the data. Experimental results on several benchmark data sets are provided to validate the effectiveness of the proposed approach. I.
Joint Clustering and Feature Selection
"... Abstract. Due to the absence of class labels, unsupervised feature se-lection is much more difficult than supervised feature selection. Tradi-tional unsupervised feature selection algorithms usually select features to preserve the structure of the data set. Inspired from the recent devel-opments on ..."
Abstract
- Add to MetaCart
(Show Context)
Abstract. Due to the absence of class labels, unsupervised feature se-lection is much more difficult than supervised feature selection. Tradi-tional unsupervised feature selection algorithms usually select features to preserve the structure of the data set. Inspired from the recent devel-opments on discriminative clustering, we propose in this paper a novel unsupervised feature selection approach via Joint Clustering and Feature Selection (JCFS). Specifically, we integrate Fisher score into the cluster-ing framework. We select those features such that the fisher criterion is maximized and the manifold structure can be best preserved simultane-ously. We also discover the connection between JCFS and other clustering and feature selection methods, such as discriminative K-means, JELSR and DCS. Experimental results on real world data sets demonstrated the effectiveness of the proposed algorithm.
