Dept. of Information and Communication

This investigation was performed to see the regularity of parameters in the modified mixture of experts (MoE) to direct parameter selection in real applications. The experiment was performed on the mean of monthly precipitation in May from 1956a through 2004a in a Chinese region. We got stable parameter combinations through testing on various combinations of parameters in various value ranges, and made statistics on the parameters in the stable combinations. The results showed that all the three parameters follow certain laws and suggest their importance. The relationship between the fuzzy weighting exponent and the value ranges of the cluster number and the index of the effectiveness factor are linear. It is of guiding significance to parameter selection in MoE systems.

With the proliferation of novel paradigms in distributed systems, including service-oriented computing, ubiquitous computing or self-organizing systems, an efficient distributed management system needs to work effectively even in face of incomplete management information, uncertain situations and dynamic changes. In this paper, Bayesian networks are proposed to model dependencies between managed objects in distributed systems management. Based on probabilistic backward inference mechanisms the so-called Strongest Dependency Route (SDR) algorithm is used to compute the set of most probable faults that may have caused an error or failure.

In real-life distributed systems, dynamic changes are unavoidable properties because of the degeneration or improvement in system performance. Hence to understand the dynamic changes and to "catch the trend" of the changes in distributed systems will be very important for distributed systems management. In order to model the dynamic changes in distributed systems, temporal extensions of Bayesian networks are employed to address the temporal factors and to model the dynamic changes of managed entities and the dependencies between them. Furthermore, the prediction capabilities are investigated by means of the relevant inference techniques when the imprecise and dynamic management information occurs in the distributed system

The level of seriousness and sophistication of recent cyber-attacks has risen dramatically over the past decade. This brings great challenges for network protection and the automatic security management. Quick and exact localization of intruder by an efficient intrusion detection system (IDS) will be great helpful to network manager. In this paper, Bayesian networks (BNs) are proposed to model the distributed intrusion detection based on the characteristic of intruders’ behaviors. An inference strategy based on BNs are developed, which can be used to track the strongest causes (attack source) and trace the strongest dependency routes among the behavior sequences of intruders. This proposed algorithm can be the foundation for further intelligent decision in distributed intrusion detection.

Overlay networks are becoming widely used for content delivery and file sharing services because they provide effective and reliable services by creating a virtual topology on top of existing networks. However, the management of overlay networks meet new challenges in the structure, security, services and deployment. This paper presents a survey of the overlay networks and correlated management. The management issues of two typical overlay networks: peer-to-peer and Virtual Private Network are investigated. As a conclusion, the management of overlay networks must import self-management (self-*) and intelligent strategies to deal with the complex management tasks, which are to deal with the dynamic, salability, resilient and security challenges on overlay networks.

When a complex information system is modelled by a Bayesian network the backward inference is normal requirement in system management. This paper proposes one inference algorithm in Bayesian networks, which can track the strongest causes and trace the strongest routes between particular effects and their causes. This proposed algorithm will become the foundation for further intelligent decision in management of information systems.

With the rising popularity of the Internet, the resulting increase in the number of available vulnerable machines, and the elevated sophistication of the malicious code itself, the detection and prevention of unknown malicious codes meet great challenges. Traditional anti-virus scanner employs static features to detect malicious executable codes and is hard to detect the unknown malicious codes effectively. We propose behavior-based dynamic heuristic analysis approach for proactive detection of unknown malicious codes. The behavior of malicious codes is identified by system calling through virtual emulation and the changes in system resources. A statistical detection model and mixture of expert (MoE) model are designed to analyze the behavior of malicious codes. The experiment results demonstrate the behavior-based proactive detection is efficient in detecting unknown malicious executable codes.

The growing complexity of distributed systems in terms of hardware components, operating system, communication and application software and the huge amount of dependencies among them have caused an increase in demand for distributed management systems. An efficient distributed management system needs to work effectively even in face of incomplete management information, uncertain situations, and dynamic changes. In this paper, Bayesian networks are proposed to model dependencies between managed objects in distributed systems. The strongest dependency route (SDR) algorithm is developed for backward inference in Bayesian networks. The SDR algorithm can track the strongest causes and trace the strongest routes between particular effects and its causes, the strongest dependency of causes can be also achieved by the algorithm. Thus, the backward inference provides an efficient mechanism in fault locating, and is beneficial for performance management.

The growing complexity of IP networks in terms of hardware components, operating system, communication and application software and the huge amount of dependencies among them have caused an increase in demand for network management systems, particularly in fault management An efficient fault detection system needs to work effectively even in face of incomplete management information, uncertain situations and dynamic changes. In this paper, dynamic Bayesian networks are proposed to model static and dynamic dependencies between managed objects in IP networks. Prediction strategies and a backward inference approach are provided for the proactive management in fault detection based on the dynamic changes of IP networks.

As networks grow in size, heterogeneity, and complexity of applications and network services, an efficient network management system needs to work effectively even in face of incomplete management information, uncertain situations and dynamic changes. We use Bayesian networks to model the network management and consider the probabilistic backward inference between the managed entities, which can track the strongest causes and trace the strongest routes between particular effects and its causes. This is the foundation for further intelligent decision of management in networks.

Recent years, requirements in network management and control have been amended by emerging network and computing models, including wireless networks, ad-hoc networks, overlay networks, Grid networks, optical networks, multimedia networks, storage networks, the convergence of next generation networks (NGN), etc. The network management has to meet resource constraints, multi-federated operations, scalability, dependability, context awareness, security, mobility, probability, etc. This paper presents a survey of the evolution of networks and their trends, the challenges of management for emerging networks and services, and the evolution of network management strategies and solutions in network management.

Bayesian networks have been used widely in modelling complex network systems. Probabilistic inferences are normal requirements in control and management of networks, particularly in modelling diagnosis systems for large and complex networks in probabilistic and dynamic environments. This paper provides a random simulation mechanism to construct the simulation in Bayesian networks for probabilistic inference, so that the simulation in Bayesian networks is close to real life networks and further the intelligent decision in management of networks can be obtained

As demands for higher data rates increase, DSL systems become incapable of keeping up due to the electromagnetic coupling present in the binders of the telephone lines. This electromagnetic coupling, known as crosstalk, is several orders greater in magnitude than the background noise. The reported techniques focusing on completely removing the crosstalk usually lead to computationally intensive solutions which may be infeasible to implement with existing hardware. For this reason, partial crosstalk removal has been proposed. In this work, we investigate the capability of the partial crosstalk cancellation for the fair or equal rate balancing among the users, subject to a constrained computational resource. Approaches which expend the computational resources in order to achieve fair or equal rates among the users were developed. Specifically, the proposed efficient Max-Min algorithm which is based on a dual optimization framework has a high convergence speed and low complexity for deployment in the xDSL systems.

Email categorization is challenging due to its sparse and noisy feature space. To address this problem, a novel semantic Vector Space Model (sVSM) using WordNet is proposed in this paper. The basic idea of sVSM is to select related semantic features that will increase the global information, and use them to enrich the semantic feature of an email. The proposed categorization method based on sVSM creates the sementic feature of an email category by both extracting terms of training email and enriching these terms with their concept-chains in WordNet. Next, tf*iwf*iwf algorithm is used to adjust the weight of the semantic feature vector. Experimental evaluations show that the proposed categorization method categorizing emails better than other email categorization methods based on traditional VSM, Baysian and KNN. More experiments show the proposed categorization method yielding better accuracy for smaller training sets with highlighting the semantic feature during identifying an email category.

Detta arbete har som syfte att studera upplevd Agency. I bakgrunden för arbetet
presenteras tidigare studier kring Agency och val i spel. Dessa används sedan för att ta
fram tre designprinciper: Tydlighet, motivation och variation. Arbetets frågeställning
gick ut på att se om dessa principer i praktiken ökade Agency och vilken princip som
var effektivast.
För att besvara frågeställningen till arbetet skapades sex stycken banor till spelet
Magnetic. Tre följde de ovan nämnda designprinciperna och tre bröt emot dessa.
Undersökningen gick sedan ut på att mäta hur dessa banor skilde sig åt kring upplevd
Agency. Undersökningen gick till så att ett antal testpersoner fick besvara en enkät
med ett påståenden kopplade till ämnet.
Resultatet visade att de scener som var skapade enligt principerna var lika bra eller
bättre än sina motparter på att skapa Agency kring just sitt eget område. Däremot
framkom det att tydlighetsprincipen och variationsprincipen påverkade varandra.
Nyckelord: Agency, val, påverkan, designprinciper