贝叶斯正则化的SOM聚类算法

发布时间：2018-05-09 19:53

本文选题：聚类 + 自组织映射(SOM)　；参考：《计算机工程与设计》2017年01期

【摘要】：研究贝叶斯正则化的自组织映射神经网络(self-organizing map,SOM)聚类训练算法。根据正则化的思想,在SOM权值调整公式中引入反映网络权值复杂性的惩罚项,避免权值调整过程中出现过度拟合。利用贝叶斯推理获取权值调整公式中的最优超参数,使迭代训练过程中网络权值和输入样本的概率分布更趋于一致,达到提升SOM聚类结果的目的。在UCI数据集上的实验结果表明,与传统的SOM算法相比,该算法的聚类凝聚度平均提升了1.5倍,聚类的准确率亦有提高,聚类效果较好。
[Abstract]:A self-organizing map neural network clustering training algorithm for Bayesian regularization is studied. According to the idea of regularization, a penalty term reflecting the complexity of network weights is introduced into the SOM weight adjustment formula to avoid over-fitting in the course of weight adjustment. By using Bayesian reasoning to obtain the optimal super-parameters in the weight adjustment formula, the network weights and the probability distribution of input samples are more consistent in the iterative training process, and the purpose of improving the SOM clustering results is achieved. The experimental results on the UCI dataset show that compared with the traditional SOM algorithm, the clustering cohesion of the algorithm is 1.5 times higher, the accuracy of clustering is also improved, and the clustering effect is better.
【作者单位】：广西大学计算机与电子信息学院;
【基金】：国家自然科学基金项目(61363027)
【分类号】：TP183;TP311.13

【相似文献】