← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8995.c · 生成模型:GPT-6

条件混合参数需满足约束

理解等级:中级

一维N分量高斯混合可由各分量均值、正方差和非负归一权重描述,网络输出原始参数后需转换到这些有效范围。分量数增加能表达更多模式,但不保证训练找到正确分布;学习目标应为最大化似然或最小化负对数似然。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      Naturally, then, we may want to predict a more expressive distribution; a mixture of two Gaussians could, for example, capture the bimodal nature of the data. An easy way to predict a distribution is to predict the parameters of some parametric family of distributions. This is precisely what we did with the Gaussian fit: the parameters of a 1D Gaussian are its mean and variance,

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第34章《条件生成模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c034.xhtml) - 结论依据:摘录中关于条件混合参数需满足约束的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:Naturally, then, we may want to predict a more expressive distribution; a mixture of two Gaussians could, for example, capture the bimodal nature of the data. An easy way to predict a distribution is to predict the parameters of some parametric family of distributions. This is precisely what we did with the Gaussian fit: the parameters of a 1D Gaussian are its mean and variance,