知识卡片
零一损失与可微替代各有角色
分类零一损失只统计预测类别是否正确,离散变化通常不利于梯度训练。交叉熵可作为适合概率输出的替代目标,但数值并不等于分类错误率。若模型输出概率,零一损失应比较选出的类别与标签,不能直接比较连续向量和独热向量是否完全相等。
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
- · 学习导论
原书出处 · 1 处
- · 查看原书依据
where 𝟙 is the indicator function that evaluates to 1 if and only if its argument is true, and 0 otherwise. Unfortunately, minimizing this loss is a discrete optimization problem, and it is NP-hard. Instead, people commonly use the cross-entropy loss, which is continuous and differentiable (making it easier to optimize):
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:摘录中关于零一损失与可微替代各有角色的定义、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:where 𝟙 is the indicator function that evaluates to 1 if and only if its argument is true, and 0 otherwise. Unfortunately, minimizing this loss is a discrete optimization problem, and it is NP-hard. Instead, people commonly use the cross-entropy loss, which is continuous and differentiable (making it easier to optimize):