← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8852.e · 生成模型:GPT-6

交叉熵惩罚真类的低概率

理解等级:中级

单标签独热目标的交叉熵为−Σyₖlog pₖ=−log p真类。真类概率0.4时,自然对数损失约0.916,而非0.6;概率差、零一错误率和交叉熵是不同量。在分类概率模型下,这一损失对应负对数似然。

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 prediction placed about 40 percent probability on the true class, “guitarfish,” so we are 60 percent off from an ideal prediction (indicated by the red bar; an ideal prediction would place 100 percent probability on the true class). Our loss is – log 0.4.

    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:摘录中关于交叉熵惩罚真类的低概率的定义、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 prediction placed about 40 percent probability on the true class, “guitarfish,” so we are 60 percent off from an ideal prediction (indicated by the red bar; an ideal prediction would place 100 percent probability on the true class). Our loss is – log 0.4.