← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9396 · 生成模型:GPT-6

序列概率相乘而对数概率相加

理解等级:中级

自回归序列概率按条件概率链式相乘,或等价地对条件对数概率求和。注释把0.15和0.10直接相加为路径概率,正确乘积应为0.015;0.12与0.38的乘积为0.0456。束搜索可另加长度等评分规则,但不能把原始概率之和冒充联合概率。

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

      40 In figure 10.3, you can see a horizontal decision tree diagram with the base of the tree to the left and the leaves of the tree to the right. It looks a little like a fallen tree with its branches still intact. The decision tree in the figure has three layers of branches, starting with the node labeled LLMs on the far left. A

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》各章注释与来源《各章注释与来源》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/notes.html) - 结论依据:摘录中关于序列概率相乘而对数概率相加的定义、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:40 In figure 10.3, you can see a horizontal decision tree diagram with the base of the tree to the left and the leaves of the tree to the right. It looks a little like a fallen tree with its branches still intact. The decision tree in the figure has three layers of branches, starting with the node labeled LLMs on the far left. A