Catálogo de publicaciones - libros

Incorporating web intelligence into website is increasingly important, especially in public sector, as the cross-section of user communities is broad. This study presents an intelligent website evolution model of public sector that can continuously provide all the targeted users with different backgrounds with the well-suited web pages to improve their satisfaction, reuse, trust and profits by continuously redesigning the current website. The model can decide what to change next in the website for the improvement of website’s outcomes of users using data mining tools and therefore the website evolves.

This paper shows how the computational model, which simulates the coordinated movements of human-like bipedal locomotion, can be evolutionally generated without the elaboration of manual coding. In the research on bio-mechanical engineering, robotics and neurophysiology, the mechanism of human bipedal walking is of major interest. It can serve as a basis for developing several applications such as computer animation and humanoid robots. Nevertheless, because of the complexity of human’s neuronal system that interacts with the body dynamics making the walking movements, much is left unknown about the control mechanism of locomotion, and researchers were looking for the optimal model of the neuronal system by extensive efforts of trial and error. In this work, genetic programming is utilized to induce the model of the neural system automatically and its effectives are shown by simulating a human bipedal gait with the obtained model. The experimental results show some promising evidence for evolutionary generation of the human-like bipedal locomotion.

Immune algorithm is a global optimal algorithms based on the biological immune theory. In this paper, a novel immune algorithm is proposed for classification rule discovery. The idea of immunity is mainly realized through two steps based on reasonably selecting vaccines, i.e., a vaccination and an immune selection. Experimental results show that immune algorithm performs better than RISE with respect to predictive accuracy and rule list mined simplicity.

This paper proposes a dynamic proportion portfolio insurance (DPPI) strategy based on the popular constant proportion portfolio insurance (CPPI) strategy. The constant multiplier in CPPI is generally regarded as the risk multiplier. Since the market changes constantly, we think that the risk multiplier should change accordingly. This research identifies factors relating to market volatility. These factors are built into equation trees by genetic programming. Experimental results show that our DPPI strategy is more profitable than traditional CPPI strategy.

Observing purely reactive situations in modern computer games, one can see that in many cases few, simple rules are sufficient to perform well in the game. In spite of this, the programming of an artificial opponent is still a hard and time consuming task in the way it is done for the most games today. In this paper we propose a system in which no direct programming of the behaviour of the opponents is necessary. Instead, rules are gained by observing human players and then evaluated and optimised by an evolutionary algorithm to optimise the behaviour. We will show that only little learning effort is required to be competitive in reactive situations. In the course of our experiments our system proved to generate better artificial players than the original ones supplied with the game.

Multiple classifier systems have shown a significant potential gain in comparison to the performance of an individual best classifier. In this paper, a weighted combination model of multiple classifier systems was presented, which took sum rule and majority vote as special cases. Particle swarm optimization (PSO), a new population-based evolutionary computation technique, was used to optimize the model. We referred the optimized model as PSO-WCM. An experimental investigation was performed on UCI data sets and encouraging results were obtained. PSO-WCM proposed in this paper is superior to other combination rules given larger data sets. It is also shown that rejection of weak classifier in the ensemble can improve classification performance further.

We present an adaptive normalization method based robust face recognition which is sufficiently insensitive to such illumination variations. The proposed method takes advantage of the concept of situation-aware construction and classifier fusion. Most previous face recognition schemes define their system structures at their design phases, and the structures are not adaptive during run-time. The proposed scheme can adapt itself to changing environment illumination by situational awareness. It processes the adaptive local histogram equalization, generates an adaptive feature vectors for constructing multiple classifiers in accordance with the identified illumination condition. The superiority of the proposed system is shown using ’Yale dataset B’, IT Lab., FERET fafb database, where face images are exposed to wide range of illumination variation.

In order to improve the generating e.ciency of the detector set, a new detection rule called edit distance rule is presented in this paper, based on the negative selection model of Arti.cial Immune System (AIS). Under this rule, edit distance is adopted to measure the similarity between self strings and randomly generated strings. Then a new detector generating algorithm used the new rule is discussed. It is necessary to use the Trie data structure to store the strings in the self set in this new algorithm. Finally, the advantages of the algorithm are given through the theoretical analysis.

Intrusion Detection based on Adaptive Resonance Theory and Artificial Immune Network Clustering (ID-ARTAINC) is proposed in this paper. First the mass data for intrusion detection are pretreated by Adaptive Resonance Theory (ART) network to form glancing description of the data and to get vaccine. The outputs of ART network are considered as initial antibodies to train an Immune Network, Last Minimal Spanning Tree is employed to perform clustering analysis and obtain characterization of normal data and abnormal data. ID-ARTAINC can deal with mass unlabeled data to distinguish between normal and anomaly and to detect unknown attacks. The computer simulations on the KDD CUP99 dataset show that ID-ARTAINC achieves higher detection rate and lower false positive rate.

A multiset of agents can mimic the evolution of the nature-inspired computations, e.g., genetic, self-organized criticality and active walker (swarm and ant intelligence) models. Since the reaction rules are inherently parallel, any number of actions can be performed cooperatively or competitively among the subsets of the agents, so that the system evolve reaches an equilibrium, a chaotic or a self-organized emergent state. Examples of natural evolution , including wasp nest construction through a probabilistic shape-grammar are provided.