A neural network ensemble is a very popular learning paradigm where the outputs of a set of separately trained neural network are combined to form one unified prediction. To improve the effectiveness of ensemble, neur...
详细信息
A neural network ensemble is a very popular learning paradigm where the outputs of a set of separately trained neural network are combined to form one unified prediction. To improve the effectiveness of ensemble, neural networks in the ensemble are not only highly correct but make their errors on different parts of the input space as well. However, most existing approaches ensemble all the available neural networks for prediction. In this paper, a selective approach to neural network ensemble based on clustering technology is presented. After individual neural networks are trained, the clustering algorithm is used to select a part of the trained individual networks in order to reduce their similarity. Then many selected neural networks are combined. Experimental results show that this approach outperforms the traditional ones that ensemble all of the individual networks.
The performance evaluation on a triple-layered LEO/MEO/GEO satellite network is carried out by Petri Net-based analysis and software simulation. The GSPN model of a triple-layered satellite network is constructed, and...
详细信息
The performance evaluation on a triple-layered LEO/MEO/GEO satellite network is carried out by Petri Net-based analysis and software simulation. The GSPN model of a triple-layered satellite network is constructed, and then used to evaluate the network performance with SPNP (Stochastic Petri nets Package) 6.0 software. The effectiveness of the results obtained by Petri Net-based analysis is verified in comp.rison with these by network simulation using OPNET, and some new results, such as the performance of the triple-layered network is better than that of LEO one under high load, are derived thereby.
How to comp.site a service from its comp.nents is a very active area of research on service-oriented architecture. This paper proposes a method of QoS-aware service comp.nents comp.sition, which has applied into Liqui...
详细信息
How to comp.site a service from its comp.nents is a very active area of research on service-oriented architecture. This paper proposes a method of QoS-aware service comp.nents comp.sition, which has applied into Liquid - a comp.nent-based embedded OS. In this method, not only functional requirements but also QoS constraints of comp.site service can be satisfied. In this approach, service is models as comp.site comp.nents that have QoS characteristics. And based on the service model and QoS model, a basic algorithm of comp.nent selection is given. In order to make this service comp.sition method suit dynamic system, heuristic and negotiation approach are used to improve the basic selection algorithm performance. At last, for the purpose of verification and comp.rison of those three different algorithms, the results of an experiment are presented.
This paper proposes a service virtualization mechanism for business end programming, with which software-level Web services can be bound to business-level services through mapping and transformation in terms of indivi...
详细信息
This paper proposes a service virtualization mechanism for business end programming, with which software-level Web services can be bound to business-level services through mapping and transformation in terms of individual business requirements. To hide software-level details and enhance the business usability of services, the paper defines a set of virtualization operations and corresponding language primitives. After introducing the design and implement of service virtualization tool, the application effect is studied through a business end programming scenario finally.
Context information can enrich services-oriented applications in multiple ways. The implementation processes of these usages differ a lot, increasing the comp.exity to develop and maintain the supporting environment. ...
详细信息
Context information can enrich services-oriented applications in multiple ways. The implementation processes of these usages differ a lot, increasing the comp.exity to develop and maintain the supporting environment. This paper proposes a context information usage pattern suitable for service comp.sition platforms. Using this pattern, a service comp.sition platform can make use of context information in a uniform way. This not only effectively supports context-aware requirements of applications at run-time but also simplifies the development and the maintenance of comp.sition platforms. The research results are evaluated in terms of experiments supported by FLAME2008 - a service mediation platform for the Olympic Games 2008.
This paper presents an approach, called TROLL, to larger-granularity service comp.sition for business-users, which is based on prefabricated and modifiable templates and constraint solving. With TROLL, business-users ...
详细信息
This paper presents an approach, called TROLL, to larger-granularity service comp.sition for business-users, which is based on prefabricated and modifiable templates and constraint solving. With TROLL, business-users can construct applications just like assembling hardware by comp.sing larger-granularity and reusable modules. This paper introduces how to use comp.sition template under temporal constraint in detail. The experimental results show that TROLL approach facilitates business-user involved service comp.sition and improve the efficiency of service-oriented application construction on business level.
A distributed cooperation intrusion detection system, which can integrate comp.ting resources of the whole network by cooperation and providing high level of introduction detection ability, is introduced. The intrusio...
详细信息
A distributed cooperation intrusion detection system, which can integrate comp.ting resources of the whole network by cooperation and providing high level of introduction detection ability, is introduced. The intrusion detection system running on each node comp.ter can prune functions which are not necessary and cooperate with others by sharing their analysis work.
Service-oriented architecture is accelerating applications of grid comp.ting from the sci.nce comp.ting to the commerce area. Most of commercial applications need the support of the transaction. This paper proposes a ...
详细信息
Service-oriented architecture is accelerating applications of grid comp.ting from the sci.nce comp.ting to the commerce area. Most of commercial applications need the support of the transaction. This paper proposes a grid transaction service (GridTS) for service grid. The GridTS can discovers dynamically grid services to perform sub-transactions, execute coordination algorithms of the atomic or cohesion transaction for these services to achieve expected consistency, according to the type of a transaction. The paper models the two types of coordination algorithms in Petri nets, and validates their correctness. The GridTS can provide powerful ability to manage transaction and hide users from the comp.ex process, provide effective supports for commercial grid applications.
The phenomenon of multiple responses to a single edge often occurs when detecting noisy ramp edges from images. This paper first investigates grayscale reconstruction for dome extraction and concludes three properties...
详细信息
The phenomenon of multiple responses to a single edge often occurs when detecting noisy ramp edges from images. This paper first investigates grayscale reconstruction for dome extraction and concludes three properties of domes extracted by it. Based on the properties, a novel approach named DEG (Dome Extraction on Gradient image) is presented to select edges in gradient images. The approach extracts h-domes in a gradient image along different directions row by row, and then selects the pixels whose value in h-domes is equal to h as edge pixel candidates. At last, edge pixels are got by thresholding the candidates. Based on the concluded properties, it can be proved that above approach has the non-maxima suppression (NMS) effect. In the gradient image, only the summits of main peaks corresponding to real edges are extracted, where as the little sharp maxima in edges' vicinity caused by noises would not be chosen. comp.tation time and memory requirements of the approach are not more than those of the non-maxima suppression approach obviously. Experiment results show that the approach suppresses the multiplicity of the response to a single ramp edge effectively without shifting edges that can be caused by smoothing methods, and gets the non-maxima suppression effect simultaneity. The approach can be used in detecting ramp edges in noisy images.
Physical skills and language skills are both fundamental intelligent abilities of human being. In this paper, we focus our attention to such sophisticated physical skills as playing sports and playing inst.uments and ...
详细信息
Physical skills and language skills are both fundamental intelligent abilities of human being. In this paper, we focus our attention to such sophisticated physical skills as playing sports and playing inst.uments and introduce research activities aiming at elucidating and verbalizing them. This research area has been launched recently. We introduce approaches from physical modeling, measurements and data analysis, cognitive sci.nce and human interface. We also discuss such issues as skill acquisition and its support systems. Furthermore, we consider a fundamental issue of individual differences occurring in every application of skill elucidation. Finally we introduce several attempts of skill elucidation in the fields of dancing, manufacturing, playing string inst.uments, sports sci.nce and medical care.
暂无评论