Research And Application Of Improved Semi-supervised Interactive Genetic Algorithm

Interactive Genetic Algorithms (IGA) which combine subjective emotional requirements ofhuman with Genetic Algorithms are the main method for solving optimization problems withimplicit objectives. Many practical problems can be recognized as this kind of optimizationproblems with implicit objectives, such as fashion design, music composition, and imageprocessing, and so on. Because these kinds of optimization problems are difficult to be quantified,we need to use Interactive Genetic Algorithms to get individual fitness by subjective humanevaluation. In the application of the advantage of subjective selection capability, we also need toface the human’s limitations when compared with machine. Human’s limitations, such ashuman’s fatigue and limitation of cognition ability often restrict the performances of IGA, for thepopulation size and the evolutionary generation are limited.This paper mainly focused on the following five aspects:(1)Describes the backgroundknowledge of interactive genetic algorithms, the basic principles and methods of existingimprovements;(2)Introduces the statistical learning theory which support vector machine basedon, the basic idea and Fundamentals of semi-supervised support vector machine;(3)Proposes anovel scheme for establishing semi-supervised support vector machine by referencing two kindsof semi-supervised kernel learning methods in the literature. And effectiveness of the algorithmis proved by experiments;(4) Present Interactive Genetic Algorithm Based on semi-supervisedSVM, which is used as the surrogate model with high generalization ability. Finally, the methodis applied to relevance feedback image retrieval, and the experimental results show that themethod is effective to improve the retrieval capabilities, achieved good search results;(5)Applies the semi-supervised interactive genetic algorithm to sensitive information monitoringsystem to generate filter template.In the fourth chapter of this paper, the kernel deformation was described in detail. Basing on estimating the geometry of the underlying marginal distribution from both labeled and unlabeleddata, we derive a data dependent kernel by incorporating the estimated geometry. The newsemi-supervised support vector machine classifier formed. These two algorithms were named"support vector machine Combined with parametric Semi-supervised Kernels" and "supportvector machine Combined with nonparametric Semi-supervised Kernels". Finally, effectivenessof the algorithm is proved by experiments.In order to alleviate user fatigue and apply the interactive Genetic Algorithm intocomplicated optimization problems, and because of the unsatisfactory performance of traditionalsurrogate models based on supervised learning, we present Interactive Genetic Algorithm basedon semi-supervised SVM. Semi-supervised SVM is used to establish the surrogate model withhigh generalization ability, and self-training method is employed for batch selecting the highreliable unlabeled samples to improve the performance of interactive genetic algorithm.Finally, the method is applied to relevance feedback image retrieval, and the experimental resultsshow that the method is effective to improve the retrieval capabilities, achieved good searchresults.In the sixth chapter of this paper, a sensitive information monitoring system is designed andimplemented. The system can filter the information flowing through the machine automatically.The system applies Interactive Genetic Algorithm based on semi-supervised support vectormachine interactive genetic algorithm in this paper to generate filter template, the process ofgenerating filter template can be controlled, and can reflect the user’s subjective needs.