Vol. 32 · No. 1

Genetic algorithm is a popular classification algorithm which creates a random population of candidate solutions and makes them to evolve into a suitable accurate solution for a given problem by processing them iteratively for several generations. During each generation the training data set is accessed by the genetic algorithm only for the population member's fitness calculation and no other extra knowledge about the problem domain is extracted from the training data set. Even the domain knowledge stored in the chromosome code of the population may be lost in the future generations due to genetic operations. All the genetic operations like crossover and mutation are probability based and they do not depend upon the domain knowledge. This phenomenon makes the genetic algorithm to converge slowly. This paper proposes a genetic algorithm which tries to gain maximum knowledge in between the generations and store them in the form of knowledge chromosomes. The gained knowledge is used to make predictions about the search space and to guide the search process to an area with potential solutions in the subsequent generations. This makes the genetic algorithm to converge quickly which in turn reduces the learning cost. The experiments show that the run time is reduced considerably when compared with the state-of-the-art evolutionary algorithm.

Given a graph G=(V, E) and k source-sink pairs (s 1, t 1), dots, (s k, t k) with each s i, t i in V, the Min-Sum Disjoint Paths problem asks to find k disjoint paths connecting all the source-sink pairs with minimized total length, while the Min-Max Disjoint Paths problem asks for k disjoint paths connecting all the source-sink pairs with minimized length of the longest path. We show that the weighted Min-Sum Disjoint Paths problem is rm FP rm NP-complete in general graphs, and the unweighted Min-Sum Disjoint Paths problem and the unweighted Min-Max Disjoint Paths problem cannot be approximated within Omega(m1- epsilon) for any constant epsilon > 0 even in planar graphs, assuming rm P neq rm NP, where m is the number of edges in G. We give for the first time a simple bicriteria approximation algorithm for the unweighted Min-Max Edge-Disjoint Paths problem and the weighted Min-Sum Edge-Disjoint Paths problem, with guaranteed approximation ratio O(łog k / łog łog k) and O(1), respectively.

Steganography in streaming media differs from steganography in images or audio files because of the continuous embedding process and the necessary synchronization of sender and receiver due to packet loss in streaming media. The conventional theoretical model for image steganography is not appropriate for explaining the security scenarios for streaming media based stegosystems. In this paper, we propose a new information-theoretical model with two pseudo-random sequences imitating the continuous embedding and synchronization characteristics of streaming media based stegosystems. We also discuss the statistical properties of Voice over Internet Protocol (VoIP) speech streams through theoretical analysis and experimental testing. The experimental results show the bit stream consisting of fixed codebook parameters in speech frames is similar in statistical characteristics to a white-noise sequence. The relative entropy between the VoIP speech stream and the embedded secret message has been found to be zero. This leads us to conclude that the proposed streaming media based stegosystem is secure against statistical detection; in other words, the statistical measures cannot detect the existence of the secret message embedded in VoIP speech streams.

A dynamic population variation (DPV) in genetic programming (GP) with four innovations is proposed for reducing computational effort and accelerating convergence during the run of GP. Firstly, we give a new stagnation phase definition and the characteristic measure for it. Secondly, we propose an exponential pivot function (EXP) in conjunction with the new stagnation phase definition. Thirdly, we propose an appropriate population variation formula for EXP. Finally, we introduce a scheme using an instruction matrix for producing new individuals to maintain diversity of the population. The efficacy of these innovations in our DPV is examined using four typical benchmark problems. Comparisons among the different characteristic measures have been conducted for regression problems and the proposed measure performed best in all characteristic measures. It is demonstrated that the proposed population variation scheme is superior to fixed and proportionate population variation schemes for sequence induction. It is proved that the new DPV has the capacity to provide solutions at a lower computational effort compared with previously proposed population variation methods and standard genetic programming in most problems.

In previous work of multiple sensing agent systems (MSASs), they mainly adjust the sensing ranges of agents by centralized heuristics; and the whole adjustment process is controlled in centralized manner. However, such method may not fit for the characteristics of MSASs where the agents are distributed and decide their activities autonomously. To solve such problem, this paper introduces the social force model for adjusting the sensing ranges of multiple sensing agents, which can make the agents adjust their sensing ranges autonomously according to their social forces to other agents and the sensing objects. Based on the social force model, the coverage and optimization models are presented for both point-type and area-type objects. The presented model can produce appropriate social forces among the sensing agents and objects in MSASs; thereby the system observability and lifetime can be improved.

The objective of this work is to improve the performance of Monte Carlo codes on Grid production infrastructures. To do so, the codes and the grid sites are characterized with simple parameters to model their behaviors. Then, a new performance model for grid infrastructures is proposed, and an algorithm that employs this information is described. This algorithm dynamically calculates the number and size of tasks to execute on each site to maximize the performance and reduce makespan. Finally, a newly developed framework called Montera is presented. Montera deals with the execution of Monte Carlo codes in an unattended way, isolating the complexity of the problem from the final user. By employing two fusion Monte Carlo codes as example cases, along with the described characterizations and scheduling algorithm, a performance improvement up to 650

Malware refers to any type of code written with the intention of harming a computer or network. The quantity of malware being produced is increasing every year and poses a serious global security threat. Hence, malware detection is a critical topic in computer security. Signature-based detection is the most widespread method used in commercial antivirus solutions. However, signature-based detection can detect malware only once the malicious executable has caused damage and has been conveniently registered and documented. Therefore, the signature-based method fails to detect obfuscated malware variants. In this paper, a new malware detection system is proposed based on information retrieval. For the representation of executables, the frequency of the appearance of opcode sequences is used. Through this architecture a malware detection system prototype is developed and evaluated in terms of performance, malware variant recall (false negative ratio) and false positives.

Contrary to existing heterogeneous data integration systems which need to be fully integrated before using, a Dataspace Support Platform is a self-sustained system which automatically provides for the user its best endeavor results regardless of how integrated its sources are. Therefore, a Dataspace Support Platform needs to support uncertainty in mediated schema and in schema mappings. This paper proposes a novel approach to automatically providing reliable mediated schemas and reliable semantic mappings in Dataspace Support Platforms. Our aim is to increase the system's endeavor results by leading it to considering as much as possible information available in any source connected. In fact, we first extract from the source schemas, their corresponding graph representations. Then, we introduce algorithms which automatically extract a set of mediated schemas from the graph representations and a set of semantic mappings between a source and a target mediated schema. Finally, we assign reliability degrees to the mediated schema generated and to the semantic mappings. Indeed, the higher the reliability degree of a given mediated schema or semantic mapping, the more consistent with the source it is. Compared with existing systems, experimental results show that our system is faster and, although completely automatic, it produces reliable mediated schemas and reliable semantic mappings which are as accurate as those produced by semi-automatic systems.

There is non-linear relationship between software metrics and defects, which results to a complex mapping. Therefore, to focus on the defect density area, it is a critical business requirement of effective and practical approach, which can help find the defect density in software releases. Soft computing provides a better platform to solve the non-linear and complex mapping problem. The aim of this paper is to formulate, build, evaluate, validate and compare two main sections of soft computing, fuzzy logic and artificial neural network approaches in prediction of defect density of subsequent software product releases. In this research, these two approaches are formulated and applied to predict the existence of a defect in file of software release. Both approaches have also been validated against various releases of two commercial software product release data sets. The validation criteria include mean absolute error, root mean square error and graphical analysis. The analysis of the study shows that artificial neural network provides better results compared to Fuzzy Inference System; but applicability of best approach depends on the data availability and the quantum of data.