知识卡片
预测方差仍可能遗漏多峰
让模型输出高斯均值和方差,可表达比单点更多的不确定性,却仍受单峰形状限制。相距很远的两个可能颜色可能被一个宽高斯连接起来,在不合理中间区域分配质量。需要更灵活分布时可考虑混合、分类或其他结构。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 条件生成模型
原书出处 · 1 处
- · 查看原书依据
We can do better by predicting a distribution rather than a point estimate. An example is shown in figure 34.1(c), where we predict a Gaussian distribution. In fact, the least-squares regression model can be described as outputting the mean of a max likelihood Gaussian fit to pθ(A | l). Predicting the distribution then just requires also predicting the variance of the Gaussian. This gives us
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:摘录中关于预测方差仍可能遗漏多峰的定义、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:We can do better by predicting a distribution rather than a point estimate. An example is shown in figure 34.1(c), where we predict a Gaussian distribution. In fact, the least-squares regression model can be described as outputting the mean of a max likelihood Gaussian fit to pθ(A | l). Predicting the distribution then just requires also predicting the variance of the Gaussian. This gives us