Abstract:Post WIMP (i,e Windows．Icons．Menus and Pointer)interface as the next generation user in terrace has suhstantial differences from WIMP Interfaee．which has been the dominont paradigm of human compu rer interaction for decades Post-WIMP interface providese more natural and efficient interaction by utilizing theadvanced interactive technology，such as virtual reality．voice technology．gesture interaction and so on How-ever，it is proved to be more difficult to be constructed In order to build post WIMP interface effectively．it is abetter way to specify the interaction at an abstract level without concerning the details of the implementation he- fore construction In this paper-the fundamental attributes of post-WIMP interface fire discussed The most dis-tinguished property of post WIMP interface is the hybrid interaction．which-ns the continuous interaetlonblending with discrete interaction The post-WIMP interface is analyzed from the point of view of hybrid system，which can give more accurate and strlet analysis for post-WIMP interface Post-WIMP interface can be modeledas a set of cooperadng hybrid automata which are special for hybrid 8yst~nl Hybrid automata are used to modelpost WlMP interface A semi formal specification language LEAFF is designed based 0n hybrid au…aton forspecifying post-WIMP interface by the combination of the text-based specification and graph-based specification，sad it can clarify the control and temporal relationships in post-WIMP interface Two typical instances，objectmanipulationin virtual reality and peninteraction，are specified byLEAFF Someissues andthefuture work are also discussed，such as the specification of paralid issues in post WIMP interaction．verification of interactionand specification based interface construction The modeling technology presented in this paper ought to be useful to the research and development of post-WIMP Imerfaee

Abstract:This paper gives an improved algorithm of λ iterate bit allocation algorithm in wavelet based image coding, which is based on the statistical characteristic of sub band block after wavelet transform. The algorithm uses the concept of λ sequence to simplify the calculation in search for optimal value and significantly speeds up the convergence process of the algorithm.

Abstract:This paper investigates the relationship between the hybrid polynomial approximation and the Hermite polynomial approximation for rational surfaces. Under some assumptions of the control point weights, the paper derives the general necessary and sufficient conditions for which both the hybrid polynomial approximation and the Hermite polynomial approximation converge to the rational surface.

Abstract:In 1975, Lander showed that there exist some languages in NP NPC P (denoted as NPI) under the assumption P≠NP. But the language constructed there is indeed an unnatural one because the construction needs to run all polynomial time Turing machines. So far, no natural problems have been proved to be in NPI under P≠NP and finding a natural problem in NP NPC P is indeed an important open problem. In this paper this long open problem is partially solved. A 2+ f(m) HSAT model is defined. Based on this model,a candidate for natural problems in ((NP-NPC)-P),denoted as NPI,under the assumption P≠NP is given,and the authors have proven that it is not in NP-Completer P≠NP.Actually,it indeed is in NPI under some stronger but plausible assumption.In comparison,similar for the two other candidates,GI and Facting,are not known.

Abstract:In this paper, based on conventional CMAC (cerebellar model architecture controller) neural network and locally weighted regression, the improved New CMAC with the same amount of memory as that of conventional CMAC is presented, which has the conventional output and its derivative information output and hence is especially appropriate for automatic control. Accordingly, the new learning algorithm is investigated, and its learning convergence is proved.

Abstract:In this paper, a robust image watermarking algorithm is proposed, which adopts the Hilbert scanning, and uses JND (just noticeable distortion) matrix based on HVS (human visual system) characteristics to mask the watermarking signal when embeds the watermark. And the strength of the watermark can be self adapting. Auto Rectification can also be added, which greatly improves the accuracy of watermark detection. Based on the description of the algorithm, experimental results and attack analysis are given. The experimental results show that this algorthm is transparent and robust.The detection results are faithful and the computational cost is low.

Abstract:During recent years, the VLSI technology has advanced profoundly, which is in urgent need of strong support from high performance integrated circuit (IC) CAD tools, that is, electrical design automation (EDA) software environment. The technology of VLSI computer aided physical design is called layout design. Global routing is a significant part of layout design. With progress in VLSI, ultra large scale integrated circuits (ULSI) deep submicron (DSM) and very deep submicron (VDSM) technology, many new problems need to be solved in global routing. To solve these problems, many methods in global routing have been proposed to improve the chip performance. Performance-Driven global routing technology has been an active area of research. As a result, the key technologies to be studied urgently in performance-driven global routing are analyzed in detail in this paper. The corresponding research work and the important results in this reserch field are also given.

Abstract:The software overhead in IN (interconnection network) communication has become currently the bottleneck of cluster system performance. In this paper, a method is proposed to reduce the communication software overhead by providing a fast message passing mechanism at user level——ULFM (user level fast message), that has been implemented in the IN TH-GBNet of cluster system. In ULFM, a new type of network access interface is provided to form a communication buffer centric application access to the cluster system IN. Several techniques, such as user level network access, zero copy message passing and simplified communication protocol, are integrated together for ULFM, that can effectively reduce the communication software overhead and achieve high communication performance with low latency and high bandwidth for application programs. Experimental results show that the latency at application level between different nodes for 16-byte message can be reduced to only 6μs, and the bandwidth can reach to 40MB/s for 2K byte length message.

Abstract:Interactive parallelizing system can provide user-friendly interactive functions and introduce user s knowledge to assist in parallelizing. It is a promising way to solve the problem of limited capability of automatic parallelizing. In this paper, an interactive environment of parallelizing compiler TIPSIE (interactive environment of Tsinghua interactive parallelizing system) is provided. Several critical techniques in constructing TIPSIE are also discussed, such as performance prediction, incremental compilation and data dependence query. The experimental results show that these techniques can get good result in improving the capability and efficiency of parallelizing compiler.

Abstract:Reducing the execution time of program is a main goal of parallel processing, while an efficient task allocation algorithm is a crucial step, especially on NOW (network of workstations). In this paper, the problem of task scheduling in structural parallel control mechanism on heterogeneous net of workstations is studied. Based on principles of greedy algorithm, granularity controlling and feeding-back-liked assignment, an algorithm on scheduling tasks is put forward under structural parallel mechanism: SSA (sub-optimal scheduling algorithm). The exper-imental results show that it can get better performance than other algorithms on heterogeneous NOW.

Abstract:It is quite difficult to compute the support vectors of massive data using the theory of support vector machine. To solve this problem, a method is brought forward to compute support vectors based on the neighborhood principle in this paper. Several questions are discussed based upon comparison and analysis of the support vector machine theory and the neighborhood principle as below: (1) The inner product function from the sample space to the dimension expand space via the feature space is constructed, and the neighborhood principle of computing the support vectors is presented; (2) Vapnik's support vector machine theory is constructed on the distance space, the algorithm is designed to compute support vectors, and the algorithm is regarded as a method to reduce the computation of quadratic programming; (3) The experimental results show that the neighborhood principle can solve the problem of support vector computation of massive data effectively.

Abstract:Multicast routing technology is important in new distributed computing such as video broadcast, video conferencing, CSCW. In this paper, a shared tree routing protocol based on distributed Genetic Algorithms (GAs) is presented, including building and dynamic maintenance of multicast routing tree in package exchange network. Computing of delay bounded Steiner Tree is also presented. With this algorithm, a dynamic constructed multicast routing tree which has a near optimal network cost under the delay bound constraint can be constructed real time. As a result, how to build and maintain a rooting tree is resolved.

Abstract:Coalition formation is a key topic in multi agent systems. People try to search for coalition structure that maximizes the sum of the values of the coalitions, but in most cases the number of coalition structures is too large to search for the optimal one exhaustively. In this paper, an algorithm is presented that within the minimal amount of search can guarantee to find a coalition structure which is within a bound from optimum. Then, the anytime algorithm searches further, and establishes a progressively lower bound, and lowers the bound rapidly.In this stage, it evidently outperforms the algorithm presented by Sandholm etc, which is the new-made better result in this area.

Abstract:The current data models used in OODBMS (object oriented database management systems) are basically divided into two categories: traditional ‘type strong’ and non traditional ‘type weak’. The feature of the former is that the objects sharing the same structure (schema) and behaviors (methods) are grouped into a class. The schema of class to which the objects belong interprets the meaning of the objects. Therefore a class must be predefined before its objects are defined. The latter is characterized by schema-less, i.e., the meaning of the objects is 'self-describing', and hence the schema need not to be defined in advance for the objects. In this paper, after outlining the main advantages and disadvantages of these two kinds of models, the authors prove that it is necessary to find some tradeoffs between traditional 'type-strong' data models and non-traditional 'type-strong' data models and non-traditional 'type-weak' ones in order to adequately accommodate the efficient and flexible manipulation of dynamic objects. For this reason, a novel dynamic data model is proposed based on CCM (conceptual clustering mechanism) which has been successfully used for the development of an object-oriented VDBMS (video database management system).

Abstract:In this paper, the state of the theory of genetic algorithms is examined, and the No Free Lunch theorem is introduced. The general framework of genetic algorithms is described, in which the performance analysis matrix is built and used to test some typical genetic algorithms. The experimental results indicate that it is a feasible method, which is easy and adaptable to the evaluation of the performance of genetic algorithms.

Abstract:TCP/IP protocol only provides services of ‘Best Effort’ level, lacking of the support of QoS. An adaptive QoS negotiation mechanism is proposed in this paper, which relies on RSVP (resource reservation protocol), carries out QoS mapping, starts the adaptability functions and the resource management functions while QoS accepting, and makes negotiate between both sides until a group of QoS values have been achieved. The basic steps of implementation by GQoS, as well as the important data structures and functions are introduced. The experimental results show that by applying the approach the QoS parameters can be adjusted adaptively when the network resources changed.

Abstract:With the simulation applications developing in extent and profundity, Internet based distributed simulation becomes the trend in the future. It faces the challenge of system scalability. Data filtering is an efficient method to improve system scalability. In this paper, after discussing the problem of system scalability, the principle, the implementation, the problems of data filtering under HLA (high level architecture) are introduced. In order to solve the problem of too much matching computation, a novel method is proposed to reduce region match computation in data filtering based on hierarchical diving of routing space. This method can reduce matching computation by suing region location knowledge to execute efficient prune searching.

Abstract:The PnP problem is a classical problem in computer vision, photogrammetry, and even in mathematics. The P5P problem is systematically investigated in this paper. It is proved algebraically that if no 3 control points among the 5 ones are collinear, the P5P problem could have at most 2 solutions, and this upper bound is also attainable. In addition, algebraic conditions are provided for the case of the unique solution and that of the two solutions of the problem respectively and a practical algorithm to compute the admissible solutions also presented. The obtained results are of practical importance in applications such as object pose estimation and robot navigation.

Abstract:The failure handling and the failure recovery are the key parts of workflow management. In this paper, a workflow model based on object is introduced. And the strategy of failure handling and failure recovery is given under the proposed workflow model. Compared with the traditional method, the new strategy takes into account both the control dependency and the data dependency between workflow tasks during the failure handling and the failure recovery. What is more, how to implement the workflow task application is explored to improve the efficiency in the failure handling and the failure recovery.

Abstract:General linear and nonlinear smoothing in an intensity image can blur corners seriously, so it's difficult to detect corners by a multi scale approach. But nonlinear multi scale CB (contour-based) morphological filters have a wonderful feature of keeping corners, at the same time it ensures that corners have the feature of monotonicity, which means non creation of new corners when the scale turns larger. Filtering the intensity image by CB morphological filters, the idea of CB morphological scale space is constructed, by which to realize multi-scale corner detection. Theory analysis and experimental results demonstrate the effectiveness of the new way.