← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8852.d · 生成模型:GPT-6

柔最大从分数构造概率

理解等级:中级

通常以pⱼ=exp(zⱼ)/Σₖexp(zₖ)把logits变成概率;给所有分数加同一常数不改变结果。原书9.19图写负号,若照用应解释为对负分数或能量做softmax,不能同时把原z称相同的未归一化对数概率。输出合法概率也不保证校准准确。

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

      The values in z are called the logits and can be interpreted as the unnormalized log probabilities of each class. Now we have, (9.20) Figure 9.9 shows what the variables look like for processing one photo of a fish during training. Figure 9.9: Softmax regression for image classification. The ⊙ symbol represents an elementwise product. The cross-entropy loss is the negative of the sum over

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第09章《学习导论》(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/c009b.xhtml) - 结论依据:摘录中关于柔最大从分数构造概率的定义、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:The values in z are called the logits and can be interpreted as the unnormalized log probabilities of each class. Now we have, (9.20) Figure 9.9 shows what the variables look like for processing one photo of a fish during training. Figure 9.9: Softmax regression for image classification. The ⊙ symbol represents an elementwise product. The cross-entropy loss is the negative of the sum over