Xufa Wang

On average time complexity of evolutionary negative selection algorithms for anomaly detection

Evolutionary Negative Selection Algorithms have been proposed and used in artificial immune system community for years. However, there are no theoretical analyses about the average time complexity of such algorithms. In this paper, the average time complexity of Evolutionary Negative Selection Algorithms for anomaly detection is studied, and the results demonstrate that its average time complexity depends on the self set very much. Some simulation experiments are done, and it is demonstrated that the theoretical results approximately agree with the experimental results. The work in this paper not only gives the average time complexity of Evolutionary Negative Selection Algorithms for the first time, but also would be helpful to understand why different immune responses (i.e. primary/cross-reactive/secondary immune response) in biological immune system have different efficiencies.

On convergence of Evolutionary Negative Selection Algorithms for anomaly detection

Evolutionary negative selection algorithms (ENSAs) are proposed by combining negative selection model and evolutionary operators. In this paper, the convergence of ENSAs with two different mutation operators is analyzed. The first mutation operator is that only one bit of a detector is selected and flipped with a high probability. The second mutation operator is that every bit of a detector has a positive probability to be flipped. The analysis results show that the ENSAs with different mutation operators have different convergent properties. Especially, the shape of the self set will affect the convergence of ENSAs with the first mutation operator.

A hybrid of differential evolution and genetic algorithm for constrained multiobjective optimization problems

Two novel schemes of selecting the current best solutions for multiobjective differential evolution are proposed in this paper. Based on the search biases strategy suggested by Runarsson and Yao, a hybrid of multiobjective differential evolution and genetic algorithm with (N+N) framework for constrained MOPs is given. And then the hybrid algorithm adopting the two schemes respectively is compared with the constrained NSGA-II on 4 benchmark functions constructed by Deb. The experimental results show that the hybrid algorithm has better performance, especially in the distribution of non-dominated set.

Infeasible elitists and stochastic ranking selection in constrained evolutionary multi-objective optimization

To handle the constrained multi-objective evolutionary optimization problems, the authors firstly analyze Debs constrained-domination principle (DCDP) and point out that it more likely stick into local optimum on these problems with two or more disconnected feasible regions. Secondly, to handle constraints in multi-objective optimization problems (MOPs), a new constraint handling strategy is proposed, which keeps infeasible elitists to act as bridges connecting disconnected feasible regions besides feasible ones during optimization and adopts stochastic ranking to balance objectives and constraints in each generation. Finally, this strategy is applied to NSGA-II, and then is compared with DCDP on six benchmark constrained MOPs. Our results demonstrate that distribution and stability of the solutions are distinctly improved on the problems with two or more disconnected feasible regions, such as CTP6.

Immune genetic programming based on register-stack structure

Inspired by biological immune principles, a novel Immune Genetic Programming based on Register-Stack structure (rs-IGP) is proposed in this paper. In rs-IGP, an antigen represents a problem to be solved, and an antibody represents a candidate solution. A flexible and efficient antibody representation based on register-stack structure is designed for rs-IGP. Three populations are adopted in rs-IGP, i.e. the common population, the elitist population and the self set. The immune genetic operators are also developed, including clone operator, recombination operator, mutation operator, hypermutation operator, crossover operator and negative selection operator. The experimental results demonstrate that rs-IGP has better performance.

Evolution of cooperation using random pairing on social networks

We studied the evolution of cooperation on social networks based on personal reputation using random pairing rule. Small-world networks and scale-free networks are used as practical network model. The iterated prisoners dilemma game are adopted as theorotical tool in which players are paired according to the network structure to play the ONE-SHOT prisoners dilemma game. Computer simulation shows that TIT-FOR-TAT-like strategy pattern will emerge from initial enviroments and cooperation can be maintained even in social networks when players have little chance to play continuous repeated games.

Hybridizing Evolutionary Negative Selection Algorithm and Local Search for Large-Scale Satisfiability Problems

This paper introduces a hybrid algorithm called as the HENSA-SAT for the large-scale Satisfiability (SAT) problems. The HENSA-SAT is the hybrid of Evolutionary Negative Selection Algorithm (ENSA), the Flip Heuristic, the BackForwardFlipHeuristic procedure and the VerticalClimbing procedure. The Negative Selection (NS) is called twice for different purposes. One is used to make the search start in as many different areas as possible. The other is used to restrict the times of calling the BackForwardFlipHeuristic for local search. The Flip Heuristic, the BackForwardFlipHeuristic procedure and the VerticalClimbing procedure are used to enhance the local search. Experiment results show that the proposed algorithm is competitive with the GASAT that is the state-of-the-art algorithm for the large-scale SAT problems.

The self-adaption strategy for parameter ε in ε-MOEA

A novel self-adaption strategy for the parameter epsiv in epsiv-MOEA is proposed in this paper based on the analyses of the relationship between the value of epsiv and the maximum number of non-dominated solutions. Then this novel strategy is applied in epsiv-MOEA and tested on 10 common benchmark functions. The experimental results demonstrate that even if without the good initial value for the parameter s, epsiv-MOEA with this self-adaption strategy (named Algorithm 1) is able to approximately obtain the expected number of non-dominated solutions, which are very close to and uniformly distributed on the Pareto-optimal front. Furthermore, the genetic drift phenomenon in Algorithm 1 is discussed Two cases of genetic drift are pointed out, and one case can be fixed up by a simple approach proposed in this paper.

Emergence of small-world networks via local interaction using prisoner’s dilemma game

The mechanism for the formation of small-world networks is important but still unsolved. We proposed a network evolution model based on local interaction among rational individuals with fixed network dimensions. This model extends Barabasis preferential attachment mechanism to consider two more realistic factors when choosing opponent to interact. Prisoners dilemma game are utilized to model such local interaction between individuals. The edges of the network are regulated by one simple rule proposed which strengthen the edges with good interaction while weaken those with bad ones. Numerical results show that small-world network structure could be evolved.

Experimental comparisons of Clonal Selection Algorithms with different metadynamics strategies

Metadynamics is an important operator of Clonal Selection Algorithms (CSAs), which is considered as an embedded mechanism to increase the population diversity. In the binary space, the traditional metadynamics usually adopts an equal probability to generate 0 and 1 for each bit of the chromosome. However, for some problems, such a metadynamics could not really increase the population diversity. In this paper, four different metadynamics strategies including the traditional metadynamics strategy and three novel metadynamics strategies are tested by experiments to compare their impacts on the performance of CSAs. The experimental results demonstrate that CSAs without and with different metadynamics strategies could have different performance.