Abstract:At present, software reliability researchers and engineers are desperately hindered by the lack of failure data. In this paper, the authors present a model, which is called JBFDM （Jade Bird failure data model）, for failure data based on the study and analysis of the relationship between software testing and reliability estimation. Furthermore, based on the model JBFDM, an approach of carrying out failure data collection and management is also proposed, and a data collection tool, which is called JBFDCollector is presented.

Abstract:There is a growing interest in the design and development of electronic commerce protocols. Accountability is the ability to hold individuals or organizations accountable for transactions. Without such assurances, electronic transactions can be susceptible to disputes. Rajashekar Kailar has proposed a framework for the analysis of communication protocols which require accountability. The authors call this framework Kailar logic. In this paper, the authors find this framework has three limitations. Firstly, the framework cannot analyze fairness of protocols. Secondly, articulating initial state assumptions is an informal and error-prone step. At last, the messages with encrypted contents in the protocol cannot be interpreted.

Abstract:The topic change of natural language is a very important clue of natural language understanding. Since different database and method should be used when different topic text is processed generally, it is important to find the topic change point in text. This technology is very useful in natural language understanding, text indexing and language model building, etc. In this paper, using the burst character of vocabulary in the change of topic, the authors present four parameters to reflect this character. They propose a method of text segmenting based on BP algorithm and hierarchical structure of word change. The accuracy of this method is about one sentence.

Abstract:Collision point is basic information to physically-based collision response, but collision point finding is a burden task, which make real-time collision response very difficult. In this paper, the authors give an algorithm to find the very fast collision time and the collision points between two objects represented by convex polygon prior to collision response.

Abstract:Parallelizing compiler can transform serial programs to parallel programs automatically or semi-automatically. The up-to-date automatic parallelizing systems cannot generate parallel codes as good as hand coding for many applications. The main reason is that they cannot manipulate the complexities of real applications. What's more, there are some semantic information in the program that the automatic tools can never know without user's knowledge. TIPS (Tsinghua interactive parallelizing system) provides some user-friendly interactive tools so that the compiler and user can cooperate with each other. It is an effective way to improve the capability and efficiency of parallelizing compiler.

Abstract:In this paper, a new BDI model for rational agents is presented by introducing assumptive belief with traditional belief in order to express the intuitive meaning of desire and intention on the cognitive aspect of the metal state of rational agents. Comparing with the BDI models from Cohen & Levesque, Rao & Georgeff, and Konolige & Pollack, this model overcomes the misunderstanding of the concepts of BDI, solves the transference problem and the side-effect problem for desire and intention, and shows both of the static and the dynamic relations between BDI, especially the maintaining and triggering role of desire.

Abstract:In this paper, an effective static task scheduling algorithm for hypercube interconnected multicomputers is presented. In the design of this algorithm, two priority lists are built for tasks and processors. During the schedule, tasks are selected from the task priority list in sequence. For a selected task, the processor on which it can be executed the earliest is assigned. In the end, the schedule results of LU decomposition and randomly generated task graphs are given and analyzed.

Abstract:Spoken Chinese analysis lies in the center of interactive speech processing system.The smallest meaningful unit in Chinese language is the word,so word sense disambiguation is the basis of spoken Chinese analysis.In this paper, the authors propose a novel method for spoken Chinese word sense disambiguation based on a simple recurrent network. This method provides a consistent processing strategy for syntax and semantics according to the internal logic between the word syntactic classification and semantic classification. Applied in the corpus for meeting schedule, this method achieves an accuracy of 96.9% in an open testing of word sense disambiguation.

Abstract:Real enterprise processes require WFMS (workflow management system) to run under heterogeneous environments involving human and automated tasks distributed across enterprises. But most of the WFMSs are based on traditional centralized DBMS, and constructed under the limitation of static organization structure. FlowAgent develops its own way to realize dynamical task-scheduling logic, and a federation structure for enterprise processes.

Abstract:To enhance the geometric ability of current modeling system, the non-regular precise geometric model is used as the unified data structure to represent and manipulate the wire-frame models, the surface models and the solid models in the algorithms. The authors present the data structure and Boolean operation in this paper. For Boolean operation, accuracy, efficiency and robustness are the main consideration. As the geometric kernel of a feature based modeling system, the algorithm has been implemented in a feature based modeling system named GS-CAD98 on NT platform.

Abstract:In this paper, the principles and the methods for correcting the nonlinear errors of the sensor system with a neural network are shown, and a novel simplified cerebella model articulation controller (SCMAC), which includes its model, algorithm and realized techniques, is proposed. The direct weight address mapping techniques are used in this model and algorithm, and the relation between the inputs and the weights is established by taking the inputs of training samples as their weight address, and the corresponding weights are found for any inputs by taking it as similar weight address, and the accurate outputs are obtained by the associable insertion algorithm. In addition, the weights are stored and addressed in a magnetic disk file, therefore, the overflow of internal memory of microcomputer are avoided, and the SCMAC is easily realized. Finally, a simulation experiment is given and the results show that the nonlinear errors of the sensor system are decreased to approximate zero after correcting with a SCMAC.

Abstract:A new fuzzy neural network model is presented in this paper. In the new model, the fuzzy rules are generated through the competitive classification of samples without the aid of domain experts. One class of samples forms a fuzzy rule. The consequent part of each fuzzy rule is a feed forward neural network trained using the corresponding class of samples. The application of the fuzzy neural network to the analysis of time series is taken as an example to demonstrate the effectiveness of the fuzzy neural network through the comparison with the conventional analysis of time series and neural network methods.

Abstract:A new unsupervised classification method using evolutionary strategies and fuzzy ART (adaptive resonance theory) neural networks is proposed in this paper. First, fuzzy ART neural networks is trained by original input samples under unsupervised way. Then evolutionary strategies is used to generate new training samples near the clusters boundaries of neural networks. Therefore the weights of fuzzy ART neural networks can be revised and refined by training those new generated samples under supervised way. The proposed method resolves the training problem for ART serial neural networks when there are only less training samples available. Consequently, it enhances the performance of ART serial neural networks and extends their application.

Abstract:There are two kinds of Bézier patches which are represented by different base functions, namely the triangular Bézier patch and the rectangular Bézier patch. In this paper, two results about these patches are obtained by employing functional compositions via shifting operators. One is the composition of a rectangular Bézier patch with a triangular Bézier function of degree 1, the other is the composition of a triangular Bézier patch with a rectangular Bézier function of degree 1×1. The control points of the resultant patch in either case are the linear convex combinations of the control points of the original patch. With the shifting operators, the respective procedure becomes concise and intuitive. The potential applications of the two results include conversions between two kinds of Bézier patches, exact representation of a trimmed surface, natural extension of original patches, etc.

Abstract:The problem of features location and label occurs frequently in the production of many types of engineering CAD. For the efficiency of features location in piping CAD system, this paper presents strategies according to the characteristics of pipe-lines in plan & section drawing. Some methods of heuristic search based on smart-line model are introduced which make it possible to accomplish automatic placement of pipe-line features in local search space.