首页> 外国专利> NUMERICAL CALCULATION METHOD FOR NON-NEGATIVE MATRIX FACTORIZATION, NUMERICAL CALCULATION DEVICE FOR NON-NEGATIVE MATRIX FACTORIZATION, PROGRAM AND STORAGE MEDIUM

NUMERICAL CALCULATION METHOD FOR NON-NEGATIVE MATRIX FACTORIZATION, NUMERICAL CALCULATION DEVICE FOR NON-NEGATIVE MATRIX FACTORIZATION, PROGRAM AND STORAGE MEDIUM

机译:非负矩阵分解的数值计算方法,非负矩阵分解的数值计算装置,程序和存储介质

摘要

PROBLEM TO BE SOLVED: To perform NMF (Non-negative Matrix Factorization) by adaptively estimating, from data comprising non-negative values (each value ≥0) such as a spectrogram of an acoustic signal or an image, a distribution shape of statistical distribution of observation data with a function defined by mixture distribution of the Poisson distribution and the Gaussian distribution as a likelihood function.;SOLUTION: Initial values of a base matrix, a mixture ratio and a weight matrix of the Poisson distribution and the Gaussian distribution are generated, the mixture ratio is updated by a Gaussian distribution rate matrix generated by dispersion of the Gaussian distribution based on the observation data, and update of the base matrix and the weight matrix is repeated until convergence is performed.;COPYRIGHT: (C)2010,JPO&INPIT
机译:解决的问题:通过从包括非负值(每个值≥ 0)的数据(例如声信号或图像的频谱图)中自适应估计数据,来执行NMF(非负矩阵分解)具有由泊松分布和高斯分布的混合分布定义的函数作为似然函数的观测数据分布;解决方案:基础矩阵的初始值,泊松分布和高斯分布的混合比和权重矩阵为生成比率,混合比率由高斯分布速率矩阵更新,该矩阵是由高斯分布的离散化基于观察数据生成的,并重复基础矩阵和权重矩阵的更新直到收敛为止。;版权所有(C)2010 ,JPO&INPIT

著录项

  • 公开/公告号JP2010039723A

    专利类型

  • 公开/公告日2010-02-18

    原文格式PDF

  • 申请/专利权人 NIPPON TELEGR & TELEPH CORP NTT;

    申请/专利号JP20080201274

  • 发明设计人 KAYANO KUNIO;KAMEOKA HIROKAZU;

    申请日2008-08-04

  • 分类号G06F17/18;

  • 国家 JP

  • 入库时间 2022-08-21 19:03:03

相似文献

  • 专利
  • 外文文献
  • 中文文献
获取专利

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号