In this paper, a power-regularization-based fuzzy clustering method is proposed for spherical data. Power regularization has not been previously applied to fuzzy clustering for spherical data. The proposed method is transformed to the conventional fuzzy clustering method, entropy-regularized fuzzy clustering for spherical data (eFCS), for a specified fuzzification parameter value. Numerical experiments on two artificial datasets reveal the properties of the proposed method. Furthermore, numerical experiments on four real datasets indicate that this method is more accurate than the conventional fuzzy clustering methods: standard fuzzy clustering for spherical data (sFCS) and eFCS.
Single Input Rule Modules connected fuzzy inference model (SIRMs model, for short) by Yubazaki et al. can decrease the number of fuzzy rules drastically in comparison with the conventional fuzzy inference models. However, it is difficult to understand the meaning of the weight for the SIRMs model because the value of the weight has no restriction in the learning rules. Therefore, the paper proposes a constrained SIRMs model in which the weights are in [0,1] by using two-phase simplex method. Moreover, it shows that the applicability of the proposed model by applying it to a medical diagnosis.
Seki et al. have proposed the functional type single input rule modules fuzzy inference model (functional-type SIRMs model, for short) which generalized consequent part of SIRMs model to function. However, it is too strict to satisfy the equivaence conditions of T–S inference model. Therefore, this paper proposes an extended functional-type SIRMs model (EF-SIRMs, for short) in which the consequent part of the functional-type SIRMs model is extended to a function with 1 dimensional polynomial from a function with n dimensional polynomial, and its properties are clarified. Further, it shows the ability of this model becomes greatly larger than that of ordinary functional-type SIRMs model. Moreover, it proposes a learning method of the EF-SIRMs model, and it is applied to a medical diagnosis, and compared with the conventional SIRMs models.
Skeleton based skin deformation methods are widely used in computer animations, with the help of some animation software, like 3D Studio Max and Maya. Most of these animation methods are based on linear blending skinning algorithm and its improved versions, showing good real-time performance. However, it is difficult for new users to use these complicated softwares to make animation. In this paper, we focus on surface based mesh deformation methods. We use spokes and rims deformation method to animate mesh models. However, this method shows poor real-time performance with high-resolution mesh models. We propose a novel animation method based on mesh decimation, making it possible to animate high-resolution mesh models in real time with the spokes and rims method. In this way, users only need to control the movement of handles to acquire intuitively reasonable animation of arbitrary mesh model. It is easier and more convenient for users to make their own animation. The experimental results show that the proposed animation method is feasible and effective and shows great real-time performance.
When employing the widely used T-S fuzzy model as a model to represent a system concerned with controller designs, it is necessary to consider the precision of the model from the point of view of control performance. Adding a term called uncertainty in the T-S fuzzy model to compensate for the difference between the concerned system and its T-S fuzzy model, this paper focuses on a design of observers for both the control state and uncertainty. Unlike a state observer in the traditional sense, which is usually designed as a whole, the state is divided into two parts by performing a unique matrix transformation; and two observers from the two divided parts of the state are designed separately in order to eliminate the influence of the uncertainty. Finally, an observer of the aforementioned uncertainty based on one of the state observers is suggested.
The flue temperature of coke oven is an important factor that guarantees the coke yield, the coke quality and the energy consumption of coking production. The heating process of coke oven is an object with multi control variables, nonlinear and large lag. The traditional PID control algorithm cannot further improve the control performance of the coke oven system. An improved implicit generalized predictive control algorithm with better control performance is proposed in this paper. Through inputting control increment value constrained by soft coefficient matrix, the calculation of matrix inversion is avoided. Soft coefficient matrix can reduce the computation time and ensure the rapidity of the system. At the same time, the input weight control law with smoothing filter is used to suppress the overshoot of the system output. Simulation results show that the proposed control method in this paper has the good control performance with faster computation speed. The proposed control method solves the problem of time variation and disturbance of coke oven system. The control algorithm of the coke oven flue temperature in this paper is effective.
In this paper, a novel modified algorithm based on MOEA/D, abbreviated as mMOEA/D, is proposed for well solving the multi-objective optimization problems. Our proposed mMOEA/D inherits from MOEA/D. In mMOEA/D, a novel elastic weight vectors design method is introduced and adopted to make those weight vectors spread more widely. On the other hand, a flexible and efficient trail DE operator is designed and used in mMOEA/D for further enhancing the performance of MOEA/D. Three groups of experimental studies are carried out. Proposed mMOEA/D is compared with the four state-the-art multi-objective optimization evolutionary algorithms on solving the multi-objective optimization problems with many objectives, and the other is that mMOEA/D is compared with MOEA/D-DE, an improved version of MOEA/D, on solving the multi-objective optimization problems with complicated PS shapes. The versions of mMOEA/D with the improvement of weight vector and DE operator are compared with MOEA/D-DE to solve multi-objective optimization problems at last. The experimental results show that mMOEA/D performs the best on almost all test instances. In other words, our proposed modification of MOEA/D is effective.
Social trading services, which are financial services connected with social networking services, are currently in the spotlight. Users can follow and automatically imitate expert traders’ trades using social trading services. Finding expert traders who exhibit an exceptional and consistent performance for users to follow is a key challenge in this field. We propose a ranking mechanism with three measures to address this issue: performance, risk, and consistency. We estimated traders’ performance, risk, and consistency levels by comparatively analyzing their trading histories and news data. In addition, we propose a system called Whom to Follow (W2F) to help users discover expert traders by utilizing this ranking mechanism. W2F visualizes the ranking results, and provides feedback functions to help users reach decisions regarding who to follow. We conducted experiments to test and then validate the proposed ranking mechanism in terms of the ranking accuracy, profit, and ranking stability. We also conducted a user experiment to demonstrate the feasibility of W2F.*
* This paper is an extended version of “Whom to Follow on Social Trading Services? A System to Support Discovering Expert Traders,” 10th Int. Conf. on Digital Information Management (ICDIM) 2015, pp. 188-193.
The Estimation of Distribution Algorithms with Graph Kernels called EDA-GK is an extension of the Estimation of Distribution Algorithms that can work with graph-related problems. Individuals of the EDA-GK are represented by graphs. In this paper, the EDA-GK is applied to solve for the Order/Degree problems, which are an NP-hard problems and are a benchmark problem in graph theory studies. Moreover, we incorporate a new sampling method for generating offspring. Experimental results on several problem instances of Order/Degree problems show the effectiveness of the EDA-GK.
The Learning Analytics is a research area that seeks to understand learning processes by using the various computer science techniques. In this paper, we focus on the analysis of certain classroom situations, such as lecturings, performing exercises, and testing. These analyses do not directory apply to students; however, they are very useful for analyzing and interpreting students’ behaviors in the classroom. This is significant as students’ behaviors can affect very real changes in classroom situations. This paper employs the Convolutional Neural Networks to identify various classroom situations from the spectrograms of environmental sounds in the classroom. Experimental results show the effectiveness of the proposed systems.
Lifestyle and genetics are known to be the major factors causing cerebral aneurysms, but some studies suggest that the shape of cerebral arteries might be correlated with the risk of aneurysm occurrence. This study focuses on the shape of cerebral arteries where cerebral aneurysms tend to occur. First, it extracts the shape feature of the cerebral artery ring, which is a predilection site of cerebral aneurysm, from 3-D magnetic resonance angiography images, and calculates four types of shape feature vectors – 3-D shape, bifurcation angle, degree of meandering, and direction of the branch points. Then, it estimates the risk of cerebral aneurysms occurring, based on the extracted features using support vector machine. To validate the proposed method, we conducted a leave-one-out cross validation test using 80 subjects (40 subjects with and 40 subjects without cerebral aneurysms). The method using a 3-D artery shape achieved 75% sensitivity and 75% specificity; the one using the bifurcation angle showed 47% sensitivity and 41% specificity. The method using the degree of meandering showed 55% sensitivity and 53% specificity, and the one that used the direction of the six branch points showed 30% sensitivity and 27% specificity. These results show that the 3-D artery shape could be a possible indicator for predicting the risk of developing cerebral aneurysms.
The configuration of solar farms, in which solar collectors are arranged in rows, is related to field and collector characteristics and solar radiation data. The main parameters considered during the optimization of solar farm designs include the number of collector rows, the center-to-center distance between collectors, collector inclination angles, and the rim angles. Solar collectors can be subjected to shading depending on the spacing between the collector rows, collector height and angle, row length, and latitude of the solar field. This study aims to optimize solar farm design by ensuring the farm receives the maximum incident solar energy and incurs the minimum deployment cost. The proposed mathematical model for photovoltaic panels is presented in detail. A multi-objective evolutionary algorithm, a non-dominated sorting genetic algorithm-II (NSGA-II), is used to achieve an optimum solar farm design that incorporates parabolic trough panels. The performances of the parabolic and flat panels are also compared, and the findings are discussed in detail. Based on the obtained results, we can verify that the parabolic PV model could generate more energy than the flat model. However, at the same cost, the flat PV model generated more energy than the parabolic model. There is a trade-off between the absolute values of the various objectives, and a solution can be selected based on the customer’s requirements and desires.
The performance of Support Vector Regression (SVR) depends heavily on its parameters, but some optimization methods based on Grid Search (GS) or evolutionary algorithms still have several issues that must be addressed. This paper proposes a new hybrid method (PSO-SS) that combines Particle Swarm Optimization (PSO) and Scatter Search (SS) to optimize the parameters of the SVR. In PSO-SS, to improve the search capability of PSO and reduce the likelihood of the PSO becoming trapped in the local optimum, the initial PSO population is generated by the diversification generation method and the improvement method of SS, and the velocity updating formula of PSO is improved by adding diversity information. On the StatLib and UCI datasets, our experiments show that the PSO-SS method is an effective parameter optimization method compared with other methods. In addition, an SVR model with its parameters optimized by PSO-SS (PSO-SS-SVR) is used to predict the grain size of aluminum alloys. The experimental results show that the PSO-SS-SVR method outperforms Back Propagation Neural Network (BPNN), PSO-SVR and the empirical model.
Unlike traditional methods that directly map different modalities into an isomorphic subspace for cross-media retrieval, this paper proposes a cross-media retrieval algorithm based on the consistency of collaborative representation (called CR-CMR). In order to measure the similarity between data coming from different modalities, CR-CMR first takes the advantage of dictionary learning techniques to obtain homogeneous collaborative representation for texts and images, then, it considers the semantic consistency of different modalities simultaneously and maps the collaborative representation coefficients into an isomorphic semantic subspace to conduct cross-media retrieval. Experimental results on three state-of-the-art datasets show that the algorithm is effective.