Abstract:In this paper, a new technique based on local preference information is proposed to improve the efficiency of Chinese parsing algorithm. By using the statistics of structure preference relations as the figures of merit, the overall efficiency of the current Chinese probabilistic parser has been improved 30% by this method, which shows good application prospects.

Abstract:The round robin servicing policies and the proportional servicing policies which can simultaneously serve for the multiple subscribers of a VOD (video-on-demand) server are studied in this paper. And how to use these policies without violating the real-time playback rates of the subscribers of a VOD server is also proposed.

Abstract:Task scheduling is an important issue in the research of parallel computing in cluster of workstations. Because it is difficult to make precise decision of task allocation when running parallel applications that dynamically spawn tasks, load imbalance maybe occur and the performance of whole system will decrease dramatically. So task reallocation is necessary for load balancing. A method for task allocation and reallocation is presented in this paper. By deferring the task's real start-time, it avoids process migration in task reallocation. Therefore, the overhead is greatly decreased. Analysis and experiments show that this method can effectively improve the performance of parallel applications in many cases.

Abstract:In underlying communication, using memory map provides user-level applications with a virtual network interface to enable user level access high speed communication devices, efficiently reduces the overhead of system protocol software and the latency in communication. In this paper, the design idea, the implementation process and the concept of communication area are discussed, by which the data can be exchanged between the kernel and user space efficiently. A sample and performance analyses are provided at last.

Abstract:In this paper, the data dependencies and lossless join decomposition in dense linear order constraint databases are studied. Several concepts of the dense linear order constraint databases are introduced. The concept of the multi-region existence dependencies is defined and studied. It is proved that the lossless join decomposition of constraint relational scheme preserves if and only if they satisfies variable independence and multi-region existence dependencies. Another dependency——partitioned multi-region existence dependencies is given. The issue of variable set under linear transformation is also proposed.

Abstract:In this paper, an agent model for the integration of multimedia object presentation is established by using events-conditions-actions mechanism and considering object-oriented characteristics of multimedia objects. This model is a dynamic model, which integrates contents, temporal relations and spatial relations of multimedia objects and supports user interactions. Because it controls objects by message passing, the encapsulations of objects can be maintained. Events-conditions-actions mechanism is employed in active database intensively and has mature technologies. The expressive ability of this model is described by using it to the temporal relations defined by Allen, and it is also verified with an example. Its contributions to multimedia database systems are also analyzed.

Abstract:Intention is an important abstract concept to specify and design agent. In this paper, an intention theory of agent computing in multi-agent system is presented to support the research on the theory of agent computing. Two intentions: achievement intention and maintenance intention, respectively denoting the different aims of agent are differentiated. The formal semantics of two intentions are defined and some important properties are obtained based on the logic framework of multi-agent system computing.

Abstract:Real-time problem solving is an interesting topic in planning in recent years. Besides discussing the deficiency of traditional planning algorithm, the authors imported the anytime algorithm, which can solve the real-time problems in this thesis. Anytime algorithm could allocate time resource reasonably to ensure the best system output performance. Anytime algorithm could be interrupted at any time and output the relatively best probable solution in that time. Genetic algorithm has the properties of the anytime algorithm. After introducing the differences between this and other search algorithms, through the experiments, the authors found that the method, which combines the random search technology and knowledge based method, could solve real-time planning problems relatively better that other methods. At last, the authors gave out the conclusion, discussed the policy of real-time planning problem solving algorithm simply, and discussed the possible developments in the future.

Abstract:In this paper, the deterministic annealing and clustering algorithms are applied to the travelling salesman problem, and a heuristic algorithm for the travelling salesman problem is put forward. The method transforms the discrete model of the travelling salesman problem into the continuous model, and the solution of the problem is obtained by solving local optimal solution of a series of problems to minimize the free energy of a physical system which varies with temperature. A simple explicit iterative formula is given. The computation results indicate that this algorithm has good performance.

Abstract:An operational HPF (high performance Fortran) compilation system is introduced in this paper. An overview of the system organization and the key techniques employed in implementing the system, such as communication detection, runtime supporting, are described in detail. To serve as concrete examples, some code fragments generated by the compiler are also included for given HPF source. Finally, some performance data are displayed, followed by a conclusion.

Abstract:ChaRM is a checkpoint-based backward fault recovery and process migration system. It is designed to recovery the fault of NOWs (networks of workstations) by checkpointing and rollback recovery. It offers functions of on-line software and hardware maintenance, process migration and load balance, etc. ChaRM is able to run on NOWs that change over time due to failure, load or availability. As long as there is at least one node alive in the cluster, the computation will complete in an efficient manner. The checkpointing, rollback recovery and process migration techniques, and some performance evaluation results are discussed in this paper.

Abstract:Although there is a spectrum of different logical definitions of model-based diagnosis, fortunately, there exists a unified abstract definition which generalizes the definitions proposed in the past. Based on this definition, the concept of kernel model-based diagnosis is proposed. Then by characterizing the process of kernel model-based diagnosis, the direct relationship between it and the prime implicants/implicates is also demonstrated. Therefore, the theoretical results in this paper can be linked with algorithms such as the ATMS (assumption-based truth maintenance system). Furthermore, it is pointed out that the characterization of kernel consistency-based diagnosis is only a special case of the characterization in this paper.

Abstract:The internal representation of objects in ODBMS (object-oriented database management system) determines the way to store and access the objects. In this paper, a kind of object representation method——RSBO (refined synchronous buffers of objects) is introduced. The method is a refinement of SBO (synchronous buffers of objects), which is a kind of object representation. With RSBO, an object having sophisticated structure is resolved into relatively simple object sectors (pieces of the object). Based on SBO, RSBO considers the mechanism of object pointer sectors, which is an expansion of object pointers. Pointer sectors not only have the similar properties of SBO sectors, but also support set-oriented computation. The access methods and possible index structures on RSBO are also discussed. The traditional index method can be still used on RSBO.

Abstract:A perfect zero knowledge proof with five moves which doesn't rely on any assumption is constructed in this paper. In this proof system, the prover can have either unlimited or limited computing power.

Abstract:In this paper, the common database interface methods on Intranet are studied and analyzed. Based on that, a new method using DB middleware under the Intranet environment is posed. The key point is to provide an efficient and powerful mechanism to manage database connection and data access to meet the needs of application requirements. It will improve the network performance for multi-user accessing database and optimize network transmission. It can provide the interface with the various RDBMSs (relational database management systems). The design considerations, the implementing and the critical techniques used are given and explained in detail in this paper.

Abstract:In this paper, a complex-feature- and MDL-based model for acquisition of VN-construction structure templates is put forward. First, a verb classification tree is created using statistical decision tree model. Then, the tree is pruned based on MDL (minimum description length) principle. Finally, structure templates are derived based on the verb classification tree. The experiments show that using the structure templates acquired with the model to recognizing VN-structure, the system has its advantages over the model based on the sense and the MLE (maximum likelihood estimation) principle in precision and recall.

Abstract:In this paper, a general method is put forward to process the induction scalars in paralleling compiler. This method changes the processing of induction scalars to the solving of difference equations and uses Z transformation and inverse Z transformation to solve equations. It improves the paralleling compiler's ability to process the inductive scalars, which is helpful to the automatic parallelization of serial programs.

Abstract:A fast and optimal parallel maximal matching algorithm is proposed for a class of graphs. It runs in O(logn) time with O((n+m)/logn) processors on a EREW PRAM (exclusive-read and exclusive-write parallel random access machine).