We propose a flexible framework for clustering hypergraph-structured data based on recently proposed random walks utilizing edge-dependent vertex weights. When incorporating edge-dependent vertex weights (EDVW), a weight is associated with each vertex-hyperedge pair, yielding a weighted incidence matrix of the hypergraph. Such weightings have been utilized in term-document representations of text data sets. We explain how random walks with EDVW serve to construct different hypergraph Laplacian matrices, and then develop a suite of clustering methods that use these incidence matrices and Laplacians for hypergraph clustering. Using several data sets from real-life applications, we compare the performance of these clustering algorithms experimentally against a variety of existing hypergraph clustering methods. We show that the proposed methods produce high-quality clusters and conclude by highlighting avenues for future work.

Jamming is a form of electronic warfare where jammers radiate interfering signals toward an enemy radar, disrupting the receiver. The conventional method for determining an effective jamming technique corresponding to a threat signal is based on the library which stores the appropriate jamming method for signal types. However, there is a limit to the use of a library when a threat signal of a new type or a threat signal that has been altered differently from existing types is received. In this paper, we study two methods of predicting the appropriate jamming technique for a received threat signal using deep learning: using a deep neural network on feature values extracted manually from the PDW list and using long short-term memory (LSTM) which takes the PDW list as input. Using training data consisting of pairs of threat signals and corresponding jamming techniques, a deep learning model is trained which outputs jamming techniques for threat signal inputs. Training data are constructed based on the information in the library, but the trained deep learning model is used to predict jamming techniques for received threat signals without using the library. The prediction performance and time complexity of two proposed methods are compared. In particular, the ability to predict jamming techniques for unknown types of radar signals which are not used in the stage of training the model is analyzed.

Driving is an essential activity in today's busy and complex society, and it demands physical and mental abilities, collectively known as a driving workload. For safe and comfortable driving, it would be useful to detect when drivers are being overloaded. Analyzing driver's workload using an electroencephalograph (EEG) is useful for this purpose. However, it is very inconvenient to obtain an EEG during actual driving, since the measuring device needs to be attached to the driver. In this paper, we develop a model to predict the driver's EEG level utilizing basic information obtained while the vehicle is being driven. We divided the EEG values into two classes, “normal” and “overload”, and extracted useful features from the vehicle driving information, such as engine RPM, vehicle speed, lane changes, and turns. A classification model using a support vector machine was built to predict normal and overload states during actual driving. We evaluated the performance of the proposed method using field-of-test data collected when driving on actual roads, and suggest directions for future research based on an analysis of the experimental results.

We present a vehicle re-identification method for a parking lot toll system. Given a probe image captured from one camera installed in the entrance of a parking lot, re-identification is the method of identifying a matching image from a gallery set constructed from different cameras in the exit region. This method is especially useful when the license plate recognition fails. Our method is based on a convolutional neural network (CNN) which is a variant of multilayer perceptron (MLP). An input image of the CNN model is cropped by the license plate detection (LPD) algorithm to eliminate the background of an original image. To train a vehicle re-identification model, we adopt the pre-trained models which showed the outstanding results in the ImageNet [1] challenge from 2014 to 2015. Then, we fine-tune one of the models (GoogLeNet [2]) for a car’s model recognition task using a large-scale car dataset [3]. This fine-tuned model is utilized as a feature extractor. Cosine function is used to measure the similarity between a probe and a gallery. To evaluate the performance of our method, we create two datasets: ETRI-VEHICLE-2016-1 and ETRI-VEHICLE2016-2. The experimental result reveals that the proposed technique can achieve promising results.

In analyzing streaming data in which the underlying data distribution may change or the concept of interest may drift over time, the ability of a classifier to adapt to drifted concepts is very important to maintaining the prediction performance. However, the true class labels of data samples are often available only after some period of time or they are obtained by experts' efforts. In this paper, we develop an effective method for active learning on data streams with concept drift. The proposed method combines active learning and adaptive incremental learning. For unlabeled data samples, the degree of concept drift is estimated and used for both data selection for labeling and adaptive incremental learning of the current classifier. Experimental results on five artificial data sets and two real data sets demonstrate a competent performance of the proposed method.

The main cause of traffic accidents is drivers' human errors such as cognitive, judgment, and execution errors. To mitigate drivers' human errors, research on the measurement and quantification of driver workload as well as the development of smart vehicles is needed. Drivers' behavior while driving includes driving straight, turning left or right, U-turns, rapid acceleration, rapid deceleration, and changing lanes. To measure and quantify a driving workload, both the subjective workload and the behavior workload caused by varied driving behaviors should be taken into account on the basis of understanding the visual, auditory, cognitive, and psychomotor characteristics of the driving workload. In this paper, we analyze electroencephalogram (EEG) data collected through an urban road driving test. To overcome large deviations of EEG values among drivers, we used EEG variation rates instead of raw EEG values. We extracted five kinds of behavior sections from the data: left-turn section, right-turn section, rapid-acceleration section, rapid-deceleration section, and lane-change section. We then selected a reference section for each of these behavior sections and compared EEG values from the behavior sections with those from the reference sections to calculate the EEG variation rates, after which we made the statistical analysis. The analysis results of this study are being used to explain the cognitive characteristics of a driving workload caused by drivers' behavior in the vehicle information system, which will provide information for safe driving by taking into account the driving workload.

We propose a user pose verification technique using depth information to compare their pose with expert riders. Xtion sensor by Asus is used for gathering a depth data in real world. The user pose verification algorithm is divided into two categories: user segmentation and user pose verification. In user segmentation step, body parts are segmented based on the region growing algorithm from a head point (i.e. seed point). Then, a simple algorithm is used to generate skeletal joints in the segmented body parts. Finally, the user pose is investigated by the standard pose of experts in the same situation.

In semi-supervised learning, when the number of data samples with class label information is very small, information from unlabeled data is utilized in the learning process. Many semi-supervised learning methods have been presented and have exhibited competent performance. Active learning also aims to overcome the shortage of labeled data by obtaining class labels for some selected unlabeled data from experts. However, the selection process for the most informative unlabeled data samples can be demanding when the search is performed over a large set of unlabeled data. In this paper, we propose a method for batch mode active learning in graph-based semi-supervised learning. Instead of acquiring class label information of one unlabeled data sample at a time, we obtain information about several data samples at once, reducing time complexity while preserving the beneficial effects of active learning. Experimental results demonstrate the improved performance of the proposed method.

Dimension reduction has been applied in various areas of pattern recognition and data mining. While a traditional dimension reduction method, Principal Component Analysis (PCA) finds projective directions to maximize the global scatter in data, Locality Preserving Projection (LPP) pursues linear dimension reduction to minimize the local scatter. However, the discriminative power by either global or local scatter optimization is not guaranteed to be effective for classification. A recently proposed method, Unsupervised Discriminant Projection (UDP) aims to minimize the local scatter among near points and maximize the global scatter of distant points at the same time. Although its performance has been proven to be comparable to other dimension reduction methods, PCA preprocessing step due to the singularity of global and local scatter matrices may degrade the performance of UDP. In this paper, we propose several algorithms to improve the performances of UDP greatly. An improved algorithm for UDP is presented which applies the Generalized Singular Value Decomposition (GSVD) to overcome singularities of scatter matrices in UDP. Two-dimensional UDP and nonlinear extension of UDP are also proposed. Extensive experimental results demonstrate superiority of the proposed algorithms.

Most traditional classifiers implicitly assume that data samples belong to at least one class among predefined several classes. However, all data patterns may not be known at the time of data collection or a new pattern can be emerging over time. In this paper, a new method is presented for monitoring the change in class distribution and detecting an emerging class. First a statistical significance test is designed which can signal for a change in class distribution. When an alarm for new class generation is set on, retrieval of new class members is performed using density estimation and entropy-based thresholding. Our experimental results demonstrate competent performances of the proposed method.

This paper presents theoretical relationships among several generalized LDA algorithms and proposes computationally efficient approaches for them utilizing the relationships. Generalized LDA algorithms are extended nonlinearly by kernel methods resulting in nonlinear discriminant analysis. Performances and computational complexities of these linear and nonlinear discriminant analysis algorithms are compared.

An optimization criterion is presented for discriminant analysis. The criterion extends the optimization criteria of the classical Linear Discriminant Analysis (LDA) through the use of the pseudoinverse when the scatter matrices are singular. It is applicable regardless of the relative sizes of the data dimension and sample size, overcoming a limitation of classical LDA. The optimization problem can be solved analytically by applying the Generalized Singular Value Decomposition (GSVD) technique. The pseudoinverse has been suggested and used for undersampled problems in the past, where the data dimension exceeds the number of data points. The criterion proposed in this paper provides a theoretical justification for this procedure. An approximation algorithm for the GSVD-based approach is also presented. It reduces the computational complexity by finding subclusters of each cluster and uses their centroids to capture the structure of each cluster. This reduced problem yields much smaller matrices to which the GSVD can be applied efficiently. Experiments on text data, with up to 7,000 dimensions, show that the approximation algorithm produces results that are close to those produced by the exact algorithm.

We propose a nonlinear feature extraction method which is based on centroids and kernel functions. The dimension reducing nonlinear transformation is obtained by implicitly mapping the input data into a feature space using a kernel function, and then finding a linear mapping based on an orthonormal basis of centroids in the feature space that maximally separates the between-class relationship. The proposed method utilizes an efficient algorithm to compute an orthonormal basis of centroids in the feature space transformed by a kernel function and achieves dramatic computational savings. The experimental results demonstrate that our method is capable of extracting nonlinear features effectively so that competitive performance of classification can be obtained in the reduced dimensional space.

A new optimization criterion for discriminant analysis is presented. The new criterion extends the optimization criteria of the classical linear discriminant analysis (LDA) by introducing the pseudo-inverse when the scatter matrices are singular. It is applicable regardless of the relative sizes of the data dimension and sample size, overcoming a limitation of the classical LDA. Recently, a new algorithm called LDA/GSVD for structure-preserving dimension reduction has been introduced, which extends the classical LDA to very high-dimensional undersampled problems by using the generalized singular value decomposition (GSVD). The solution from the LDA/GSVD algorithm is a special case of the solution for our generalized criterion, which is also based on GSVD. We also present an approximate solution for our GSVD-based solution, which reduces computational complexity by finding subclusters of each cluster, and using their centroids to capture the structure of each cluster. This reduced problem yields much smaller matrices of which the GSVD can be applied efficiently. Experiments on text data, with up to 7000 dimensions, show that the approximation algorithm produces results that are close to those produced by the exact algorithm.