The Study Of Semi-supervised Learning Model Based On Artificial Immune System

As an important field of study in pattern recognition and machine learning, the semi-supervised learning algorithm has been developed in theory and practice for a long time along with the applications of machine learning in data analysis and data mining problems. The semi-supervised learning is an algorithm that study with labeled and unlabeled samples at the same time in the condition of knowing part of labeled samples. Comparing with the supervised learning, it saves the cost of tagging samples. Comparing with the unsupervised learning, it makes sure the accuracy.The additional semi-supervised learning is always on the basis of complicated mathematical formula, and it's very complicated while operated. In recent years, bionic mechanism based on biological systems is increasingly becoming research focus in many fields. Such as genetic algorithm, ant colony optimization and particle swarm optimization. They are the basic principles of biological research, and gradually be applied to machine learning, pattern recognition and so on. By applying some principles of biological systems to the areas of artificial intelligence, you can achieve the desired effects of simplifying the calculation and operation and improving the accuracy.The papers proposed a new semi-supervised learning algorithm according to the biological immune mechanism. The new algorithm applies basic principles of immune system, such as antigen-antibody, immune response, immune memory and so on, to the semi-supervised learning and take the initial response as training phase, secondary response as testing phase. By learning the training samples through evolutionary learning algorithm, we make full use of the powerful information processing capability of immune system and overcome the shortcomings of traditional semi-supervised learning algorithm. Finally, we verify the feasibility of the algorithm by testing on face images of ORL. The results of test show that the new algorithm has a good capacity of self-learning, self-organizing and adaptive. Comparing with the supervised learning algorithm based on artificial immune system, it saves the cost of marking samples, and achieves very good results in recognition rate.