基于信息叠加的学习算法

LEARNING ALGORITHM BASED ON THE SUPERPOSITION OF INFORMATION

提出了一种基于信息叠加的迭代学习算法 .该算法利用协同神经网络中的原型模式具有信息的可加性 ,将学习中误识率最高的模式作为反馈量来修正原型模式 .利用实际采集得到的样本对新算法进行的测试表明 :新算法具有最优搜索能力强 ,训练时间短的特点 .另外 ,将新算法与基于遗传算法的原型模式选取算法在网络训练性能上进行了比较 .

A new iterative learning algorithm based on the superposition of information was proposed. Because the prototype patterns of synergetic neural network (SNN) has the ability of superposition of information, the new algorithm can modify the prototype patterns using the pattern, of which the recognition rate is the lowest during training as the feedback. The test upon the samples from real environment shows that the new algorithm has the characteristic of strong ability of optimal searching and the shorteness of training time. Additionally, the comparison of training performance between the new algorithm and selection algorithm of prototype patterns based on genetic algorithm(SAPPGA) was made.