Earticle

Support vector machine (SVM) is a kind of machine learning method, but the selection of parameters has important effects on the generalization ability of SVMs. In this study, the relation between the error penalty parameter C, kernel parameter σ and the generalization ability of SVMs is discussed. Parameter C adjusts the similarity among within-class members, while parameter σ adjusts the similarity between classes. Moreover, C and σ balances each other mutually within a certain range, which forms a fan-shaped optional parameter distribution region. The optimal parameter area should be located near the center of the sector where both C and σ are small. According to this, a method is suggested to first search a suitable area with coarse grids, and then determine the optimal parameter within the area with a fine bilinear grid. Experimental results show that the new parameter selection method can not only avoid local optima, and thus excluding the cases in which C and σ are big and unstable, but also can be extremely fast in searching process. Compared with other parameter selection methods, the performance of SVMs cannot be influenced, or even better in some cases.

Clustering algorithm is a common analysis technology for network forensics, which, lacking of any prior knowledge, can effectively find out the invasions by analyzing the collected real-time communication data flowing through the network. This paper proposed an improved dynamic kernel clustering algorithm for mixed numeric and categorical network communication data. First, centroid prototype based on the mean and distribution centroid was put forward to represent the cluster center. Then by using Gaussian kernel function, the paper introduced a new dissimilarity measure between the data object and the centroid prototype in combination with the significance of different categorical values. On this basis, the objective function was defined, which took into account both the compact degree in a cluster and the discrete degree among the clusters. After that an improved kernel clustering algorithm was designed. In the process of clustering, centroid prototype and the value of the clustering parameter dynamically updated for a better description of the characteristics of clusters’ change. Finally, in order to verify the feasibility and effectiveness of the algorithm, the paper further applied it to network forensics, and the experimental results showed that the method could mine the intrusion behavior more accurately.

The Project presents object tracking from videos based on template matching using Gaussian kernel with probability distribution function and mean shift algorithm. Initially, the target will be selected from chosen video sequence to track desired object in consecutive frames. The target will be utilized to determine the probability distribution function for similarity measurement between target and current processing frames. Here, Gaussian kernel function and its gradient are used here to find the PDF for corresponding templates. Similarity between two different images will be measured by weighted sum of Gaussian coefficients and PDFS. Mean shift approach used here to shifting the starting coordinates of template to find its similar features in consecutive frames to detect desired objects. The dissimilarity between the target model and target candidates will be expressed by a metric derived from Bhattacharyya coefficient. The project simulated results shows that moving object from video will be tracked accurately at different position and shape with help of templates in a considerable amount of time.

Clustering is one of the main tasks used in pattern recognition and classification. Out of many methods that have been reported till date the most widely used methods are based on likelihood approach of mixture model. Among different mixture models, Expectation Maximization for Gaussian Mixture is most exploited and trusted algorithm for data clustering. However, it has some short comings such as initial parameters are to be given a-priori, convergence speed is slow and the results obtained are highly dependent upon the initial parameters. Many variations have been carried out in implementing EM algorithm but still there is ample scope for improvement. The proposed algorithm tries to overcome these shortcomings and provide more robust and efficient version of clustering algorithm. An improvement related to cluster partitioning is proposed in the existing algorithm resulting some advantages. The robustness and efficacy of the algorithm is demonstrated qualitatively as well as quantitatively with the help of some experiments.

Rapid and timely monitoring of traumatic inflammation is conducive to doctors’ diagnosis and treatment. It has been proved that electronic nose (E-nose) is an effective way to predict the bacterial class of wound infection by smelling the odor produced by the metabolites, and the classification accuracy of E-nose is influenced strongly by the classifier. To improve the performance of E-nose in predicting the bacterial class of wound infection, an enhanced SVM with a novel weighted Gaussian RBF kernel is proposed in this paper, and the way of setting parameters of this enhanced SVM is also given. Experimental results prove that the classification accuracy of SVM with the novel weighted Gaussian RBF kernel is 95.24%, which is better than other considered classifiers (PLS-DA, RBFNN, SVM with single Gaussian RBF kernel and SVM with traditional weighted Gaussian RBF kernel). All results make it clear that the enhanced SVM proposed in this paper is an ideal classifier when E-nose is used to detect the bacterial class of wound infection.

Most hyper-ellipsoidal clustering (HEC) approaches use the Mahalanobis distance as a distance metric. It has been proven that HEC, under this condition, cannot be realized since the cost function of partitional clustering is a constant. We demonstrate that HEC with a modified Gaussian kernel metric can be interpreted as a problem of finding condensed ellipsoidal clusters (with respect to the volumes and densities of the clusters) and propose a practical HEC algorithm that is able to efficiently handle clusters that are ellipsoidal in shape and that are of different size and density. We then try to refine the HEC algorithm by utilizing ellipsoids defined on the kernel feature space to deal with more complex-shaped clusters. The proposed methods lead to a significant improvement in the clustering results over K-means algorithm, fuzzy C-means algorithm, GMM-EM algorithm, and HEC algorithm based on minimum-volume ellipsoids using Mahalanobis distance.

To improve the ability to determine a vehicle's movement information even in a challenging environment, a hybrid approach called non-Gaussian square rootunscented particle filtering (nGSR-UPF) is presented. This approach combines a square root-unscented Kalman filter (SR-UKF) and a particle filter (PF) to determinate the vehicle state where measurement noises are taken as a finite Gaussian kernel mixture and are approximated using a sparse Gaussian kernel density estimation method. During an outage-free GPS period, the updated mean and covariance, computed using SR-UKF, are estimated based on a GPS observation update. During a complete GPS outage, nGSR-UPF operates in prediction mode. Indeed, because the inertial sensors used suffer from a large drift in this case, SR-UKF-based importance density is then responsible for shifting the weighted particles toward the high-likelihood regions to improve the accuracy of the vehicle state. The proposed method is compared with some existing estimation methods and the experiment results prove that nGSR-UPF is the most accurate during both outage-free and complete-outage GPS periods.

Consider a nonlinear transform ${\Phi}(x)$ of x in $\mathbb{R}^p$ to Hilbert space H and assume that the dot product between ${\Phi}(x)$ and ${\Phi}(x^{\prime})$ in H is given by < ${\Phi}(x)$, ${\Phi}(x^{\prime})$ >= K(x, x'). The aim of this paper is to propose a mathematical technique to take screen shots of the multivariate dataset mapped to Hilbert space H, particularly suited to Gaussian kernel $K({\cdot},{\cdot})$, which is defined by $K(x,x^{\prime})={\exp}(-{\sigma}{\parallel}x-x^{\prime}{\parallel}^2)$, ${\sigma}$ > 0. Several numerical examples are given.

본 논문에서는 비선형 데이터를 구분하기 위하여 비선형 Support Vector Machine (SVM)을 고려한다. 비선형 SVM의 성능은 사용되는 커널 함수 및 이를 구성하는 파라미터에 의하여 크게 좌우된다. 대부분의 기존연구들이 이러한 파라미터들을 데이터를 통하여 단일 추정하여 사용하였고, 이는 분류기의 확장성 및 불확실성 표현력을 저하시키는 주된 요인으로 여겨졌다. 이에, 본 연구에서는 커널함수의 파라미터를 Type 2 Fuzzy Number (T2FN)으로 표현하고 이를 사용하는 비선형 SVM 프레임웍을 제시한다. 제시된 프레임웍은 비선형 SVM의 불확실성 표현력을 높여줌과 동시에, 데이터에 맞는 비퍼지화 알고리즘을 통하여 도출되는 결과의 객관성을 높이는 장점을 지닌다. 이를 증명하기 위한 수치모델로서, 가우시안 커널 타입의 비선형 SVM을 고려하고, 커널 함수의 파라미터를 T2FN으로 추정하여 설정한 입력데이터를 분류한 뒤 이를 기존 알고리즘과 비교한다. 제시된 프레임웍 및 내재된 알고리즘은 데이터를 통해 T2FN을 설정하고, 이를 통해 확장성 있는 비선형 커널 기반 SVM을 설계하는데 기여한다.

In order to classify non-linearly aligned data, this paper considers an effective nonlinear support vector machine (SVM). The performances of a nonlinear SVM depends upon the used kernel functions and their parameters. Most of the existing research studies have estimated these parameters using point based estimations, these tendencies are considered as main obstacles decreasing the embedding uncertainties and the machine's extension-ability. In order to overcome these limitations, this research uses Type 2 Fuzzy Number (T2FN) and T2FN embedding nonlinear SVM framework. The proposed SVM framework incorporates more extended uncertainties as well as helps the robust data analyses using the suggested defuzzification algorithm. In order to show the effectiveness of the framework, numerical examples with Gaussian kernel function based nonlinear SVM are considered. Parameters in the used kernel function are estimated using T2FN and the classifications using the provided algorithm are analyzed. The provided framework and estimation methods contribute to the design of nonlinear SVM possessing more expandable uncertainties and analyzing abilities using T2FN.

GPS 수신이 어려운 실내 환경에서 이동 단말기 사용자나 로봇의 위치를 추정하기 위해 WiFi 신호 강도를 이용하는 연구가 최근 들어 활발히 진행되고 있다. 본 논문에서는 WiFi 신호의 불안정성과 불확실성에 효과적이고 이동 단말기에서 계산 성능을 고려하여 가우시안 프로세서를 변형한 방법을 적용하여, 실내 연속 공간에서 이동 중인 스마트폰 사용자의 실시간 위치를 추정하는 방법을 제안한다. 실험을 통해 제안한 방법의 성능을 분석해보고, 성능 개선을 위한 확장 방안을 제시한다.

본 논문에서는 비지도학습 모델인 오토인코더와 가우시안 커널 밀도 추정 함수를 이용하여 차량용 CAN 네트워크에서 비정상적인 데이터를 탐지하는 방안을 제안한다. 제안하는 오토인코더 모델은 정상 데이터에서 CAN 프레임의 ID만으로 학습시킨다. 이후 가우시안 커널 밀도 추정 함수를 이용하여 구한 최적의 프레임 개수와 손실 임계값을 가지는 모델을 사용하여 비정상 데이터를 효과적으로 탐지한다. DoS 공격, Gear 스푸핑 공격, RPM 스푸핑 공격, Fuzzy 공격 등 4가지 공격 데이터로 오토인코더 기반 IDS를 검증하였으며 성능을 평가하였다. 기존 비지도학습 기반 모델들과 비교했을 때 우수한 성능을 나타냈으며 모든 평가 지표에서 99% 이상의 성능을 나타냈다.

This paper proposes an approach to detect abnormal data in automotive controller area network (CAN) using an unsupervised learning model, i.e. autoencoder and Gaussian kernel density estimation function. The proposed autoencoder model is trained with only message ID of CAN data frames. Afterwards, by employing the Gaussian kernel density estimation function, it effectively detects abnormal data based on the trained model characterized by the optimally determined number of frames and a loss threshold. It was verified and evaluated using four types of attack data, i.e. DoS attacks, gear spoofing attacks, RPM spoofing attacks, and fuzzy attacks. Compared with conventional unsupervised learning-based models, it has achieved over 99% detection performance across all evaluation metrics.