基于记忆的SVM相关反馈算法

Relevance Feedback Algorithm Based on Memory Support Vector Machines

支持向量机(SVM)方法并不假设样本的分布条件,它基于结构风险最小化原则,对小样本情况下的学习问题给出最优解,并且在样本趋于无穷时能保持良好的一致收敛性。在SVM的基础上提出的MSVM方法,通过记忆功能,用历次反馈的累积样本代替一次反馈样本,从而增加了学习样本数量,减小了查准率的振荡,提高了检索精度;同时为了减轻用户负担,提出了记忆性标注。实验证明,MSVM方法可以避免因训练样本集过小而出现的局部最小化的问题,能较为准确地分类图像库中的图像,同时有效地减轻了用户的负担。

Support vector machine（SVM） is based on the minimum of structure risk and used for small samples in machine learning.Memory support vector machine（MSVM） feedback is based on SVM and used cumulation samples replacing feedback samples by memory.It reduces the risk of recall vibration.MSVM feedback also proposes memory label which is used for lightening user＇s burden.MSVM feedback is proved its superiority by relevant experiments.