基于统计学习理论优化感知器的遗传算法

GENETIC ALGORITHM OF OPTIMIZING PERCEPTION BASED ON STATISTICAL LEARNING THEORY

本文提出一种基于统计学习理论优化感知器的遗传方法。该方法将遗传算法和神经网络相结合,通过统计学习理论指导遗传算法优化分类器的过程,避免了传统的感知器分类的偏向性、连接权的局部收敛性、误识率高等弱点;借助于遗传算法全局寻优的特点,使改进后的算法,具有自进化、自适应能力,以及很好的数据推广性能和抗干扰性,提高了神经网络的整体性能,与标准的SVM算法相比,具有更广阔的应用范围。

In this paper, we demonstrate a genetic algorithm of optimizing perception based on statistical learning theory. The method combines genetic algorithm with neural network and uses statistical learning theory to direct the genetic algorithm to optimize perception, to avoid deflection of traditional perception, local convergence of connection weight and high probability of failed recognition. Owing to the global optimization of genetic algorithm, the improved method can evolve and adapt itself. It has better generalization performance and better property of avoiding disturbance, which improves the total performance of neural network. Comparing with normal algorithm of SVM, it has wider range of applications.