首页> 外国专利> Standard pattern learning device, labeling standards calculating device, standard pattern learning method and program

Standard pattern learning device, labeling standards calculating device, standard pattern learning method and program

机译:标准图案学习装置,标签标准计算装置,标准图案学习方法及程序

摘要

PPROBLEM TO BE SOLVED: To improve the performance of an identification model in integrated learning of the identification model and a generation model through the use of labelled learning data and unlabelled learning data. PSOLUTION: A standard pattern learning device includes: an identification generation model integrated learning means 500 for integrally learning the identification model and the generation model under a restriction including a penalty item with respect to discrepancy between the parameters of the two models through the use of the labelled learning data and the unlabelled learning data; and a labeling criterion calculating means 501 for performing class identification processing by each of the identification model and the generation model learned by the identification generation model integrated learning means 500 with respect to each kind of unlabelled learning data to be given to the identification generation model integrated learning means 500, and calculating a labeling criterion indicating the necessity of adding an additional label, based on a difference in the result of identification between the identification model and the generation model. PCOPYRIGHT: (C)2011,JPO&INPIT
机译:

要解决的问题:通过使用标记的学习数据和未标记的学习数据,在识别模型和生成模型的集成学习中提高识别模型的性能。解决方案:一种标准模式学习设备包括:识别生成模型集成学习装置500,用于在包括关于两个模型的参数之间的差异的惩罚项的限制下,整体学习识别模型和生成模型。标记的学习数据和未标记的学习数据的使用;标记标准计算装置501,用于对要提供给集成的识别生成模型的各种未标记的学习数据,通过由识别生成模型综合学习装置500学习的识别模型和生成模型分别进行类别识别处理。学习装置500,并基于识别模型和生成模型之间的识别结果的差异,计算指示添加附加标签的必要性的标签标准。

版权:(C)2011,日本特许厅&INPIT

著录项

  • 公开/公告号JP5387274B2

    专利类型

  • 公开/公告日2014-01-15

    原文格式PDF

  • 申请/专利权人 日本電気株式会社;

    申请/专利号JP20090217804

  • 发明设计人 大西 祥史;

    申请日2009-09-18

  • 分类号G06N5/04;G10L15/06;G10L15/10;G06T7/00;

  • 国家 JP

  • 入库时间 2022-08-21 16:13:57

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