生成模型(Generative)和判别模型(Discriminative)
- 引言
最近看文章《A survey of appearance models in visual object tracking》(XiLi,ACMTIST,2013),在文章的第4节第1段有这样的描述,“Recently,visualobject tracking has been posed as a tracking-by-detection problem, where statistical modeling is dynamically performed to support object detection. According to the model-construction mechanism, statistical modeling is classified into three categories, including generative, discriminative, and hybrid generative-discriminative.”随后又再以前看的《Fast Compressive Tracking》(Kaihua Zhang,PAMI,2014)的第2节第1段找到相应的话,“Recent surveys of object tracking can be found in [22]-[24],In this section, we briefly review the most relevant literature of on-line object tracking. In general,tracking algorithms can be categorized as either generative ordiscriminative based on their appearance models.”类似还有很多,每次看到都会有点模糊,感觉心中没底,所以就找了些资料总结了下,有不对的地方还请大家指正。
2.
概念

本文探讨生成模型(Generative)和判别模型(Discriminative)在监督学习中的差异。生成模型学习数据的联合分布,能反映数据相似度,但不关注分类边界;判别模型直接学习决策函数,追求高准确率,适用于分类问题。举例说明中,生成模型类似于学习所有语言来识别语音,而判别模型仅学习语言之间的差异。在跟踪算法中,生成模型用于模式匹配,判别模型寻找决策边界。
最低0.47元/天 解锁文章
4608

被折叠的 条评论
为什么被折叠?



