纺织高校基础科学学报2018，Vol.31Issue(1)：122-127,6.DOI:10.13338/j.issn.1006-8341.2018.01.020
基于L1范数改进的自回归算法及分类应用
An improved autoregressive algorithm based on L1norm and classification application
摘要
Abstract
In order to further improve the neural autoregressive density estimation algorithm to fit the joint probability distribution ability and increase the classification accuracy,a neural au-toregressive density estimation algorithm based on regularization of L1 norm parameters is pro-posed,and the Polyak averaging parameter updating ideas is added based on L1 regularization. The network sparsity processing and connection weight of the stable updating improve the ac-curacy of classification.By adjusting the super-parameter,the UCI data set is selected to test the fitting ability of probability distribution of the improved algorithm,the result show sthat the fitting ability is improved.By testing the classification accuracy rate of the image data set,the classification accuracy rate of data set LabelMe increased from 83.43% to 83.85%,the classification accuracy rate of data set UIUC-Sports increased from 77.29% to 78.12%,the experiment results indicate that the improved algorithm is effective.关键词
自回归密度估计/正则化/波利亚科夫平均/图片分类Key words
autoregressive density estimation/regularization/Polyak averaging/picture classi-fication分类
数理科学引用本文复制引用
陈国泽,贺兴时..基于L1范数改进的自回归算法及分类应用[J].纺织高校基础科学学报,2018,31(1):122-127,6.基金项目
陕西省自然科学基金(2016JQ1022) （2016JQ1022）
西安市教育重大招标项目(2015ZB-ZY04) （2015ZB-ZY04）
