知识卡片
分类分布表达力受离散空间限制
k类概率向量可表示给定k个离散值上的任意概率质量分布,因此能容纳多个模式。但这不消除量化误差,也不代表线性Softmax条件映射能拟合任意输入关系。需要区分输出分布族的表达力与参数网络的表达力。
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
One of the most common choices for an expressive, multimodal family of distributions is the categorical distribution, Cat. This distribution applies only to discrete random variables, so to use it, the first thing we have to do is quantize our possible output values. Figure 34.1(d) shows an example, where we have quantized the a-value into ten bins. Then the categorical distribution is simply a ten-dimensional
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:摘录中关于分类分布表达力受离散空间限制的定义、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:One of the most common choices for an expressive, multimodal family of distributions is the categorical distribution, Cat. This distribution applies only to discrete random variables, so to use it, the first thing we have to do is quantize our possible output values. Figure 34.1(d) shows an example, where we have quantized the a-value into ten bins. Then the categorical distribution is simply a ten-dimensional