知识卡片
密度无偏不推出对数无偏
样本平均可以无偏估计正的边缘密度,取对数后一般因凹性产生向下偏差。它也不等于样本对数密度的平均。更大样本数可能在适当条件下改善一致性,但不能把有限样本目标、真实对数似然及其最优参数直接混同。
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
原书出处 · 1 处
- · 查看原书依据
We now have a differentiable objective for both ψ and θ; the objective for ψ is expressed as an expectation and can be optimized by taking a Monte Carlo sample from that expectation. We could also try using Monte Carlo to approximate the objective for θ, but this would yield a biased estimator of θ, since equation (33.24) has the log outside the expectation.
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第33章《生成建模与表示学习》(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/c033.xhtml)
- 结论依据:摘录中关于密度无偏不推出对数无偏的定义、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 now have a differentiable objective for both ψ and θ; the objective for ψ is expressed as an expectation and can be optimized by taking a Monte Carlo sample from that expectation. We could also try using Monte Carlo to approximate the objective for θ, but this would yield a biased estimator of θ, since equation (33.24) has the log outside the expectation.