Publications

This article elaborates on the role of neural networks in data mining, especially classification, and presents various ways of using them in this area. In order to do this the main architectures of neural networks (including multilayer perceptrons and radial basis function networks) are reviewed, and an overview of the classification process and the training of neural networks is given. Furthermore, the interpretation of neural networks and the generation of rules based on already trained networks are discussed and exemplified on a number of rule extraction algorithms. Finally, the role of neuro-fuzzy systems in the process of designing interpretable neural networks is described.

In this paper, a robust gain-scheduled path following controller for automated vehicles based on learning methods is presented. Two major challenges are overcome:1) Varying longitudinal velocity, uncertain cornering stiffness, and unmodelled uncertainties make dynamic-model-based controller design work complex. 2) Driving scenario changes deteriorate path following controller performance. An effective learning method, online updating least squares-support vector machine (LS-SVM) model is adopted for vehicle path following system considering varying velocity and cornering stiffness in this paper. Then the updating LS-SVM model is transformed into linear-parameter-varying (LPV) model with disturbance. The robust <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$H_\infty$</tex-math></inline-formula> controller design method is novelly employed to design path following controller for updating LS-SVM model. By this method a gain-scheduled output-feedback controller is designed. To improve transient performance, the poles of closed-loop system are assigned to desired regions. Simulation results using a high-fidelity and full-car model from CarSim have verified the effectiveness of the proposed control strategy.

Fuzzy cognitive maps (FCMs) form an important class of models for describing and simulating the behavior of dynamic systems through causal reasoning. Owing to their abilities to make the symbolic knowledge processing simple and transparent, FCMs have been successfully used to model the behavior of complex systems originating from numerous application areas, such as economy, politics, medicine, and engineering. However, the design of FCMs necessarily involves domain experts to develop a graph-based model composed of a collection of system's concepts and causal relationships among them. Consequently, since humans exhibit an intrinsic factor of subjectivity and are only able to efficiently develop small-size graph-based models, there is a legitimate need to devise methods capable of automatically learning FCM models from data. This research addresses this need by introducing a competent memetic algorithm to generate FCM models from available historical data, with no human intervention. Extensive benchmarking tests performed on both synthetic and real-world data quantify the performance of the competent memetic method and emphasize its suitability over the models obtained by conventional and noncompetent hybrid evolutionary approaches in terms of accuracy, approximation ability, and convergence speed. Moreover, the proposed approach is shown to be scalable due to its capability to efficiently learn high-dimensional FCM models.

Pair programming is one of the most controversial parts of XP. Claims are mostly based on anecdotal evidence and limited experimentation performed in classroom settings. This paper reports the preliminary results of an analysis of the effects of pair programming on job satisfaction. A questionnaire on pair programming techniques has been compiled and posted on the web. 108 responses have been collected from around the world. The preliminary results evidence a very positive effect of pair programming on job satisfaction.

Fuzzy C-Means (FCM) is a widely used clustering method. However, FCM and its many accelerated variants have low efficiency in the mid-to-late stage of the clustering process. In this stage, all samples are involved in updating their non-affinity centers, and the membership grades of most samples, whose assignments remain unchanged, are still updated by calculating the sample-center distances. All these factors lead to the algorithms converging slowly. In this paper, a new affinity filtering technique is developed to recognize a complete set of non-affinity centers for each sample with low computations. Then, a new membership scaling technique is suggested to set the membership grades between each sample and its non-affinity centers to 0 and maintain the fuzzy membership grades for others. By integrating these two techniques, FCM based on new affinity filtering and membership scaling (AMFCM) is proposed to accelerate the whole convergence process of FCM. Numerous experimental results performed on synthetic and real-world data sets have shown the feasibility and efficiency of the proposed algorithm. Compared with state-of-the-art algorithms, AMFCM is significantly faster and more effective. For example, AMFCM reduces the number of FCM iterations by 80 <inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> on average.