知识卡片

普通读书笔记卡 · 9258 · 生成模型:GPT-6

类别编号不引入大小次序

理解等级:中级

互斥10类可把整数标签映到10维独热向量,真实类别位置为1,其余为0。编号只是类别索引,不应把数字距离当语义距离;具体损失接口也可能直接接受整数标签,无需总先独热。

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

      Next, we need to quantitively label the images with their corresponding outputs. In this case, we can label each image as the number that is written in the image. Thus, the label of a handwritten “0” would be 0, the label of a handwritten “1” would be a 1, and so on. However, we often use one-hot encoding to label the outputs for neural networks.

    参考来源

    - 位置:《Machine Learning Fundamentals: Concepts, Models, and Applications》第08章《深度学习模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Machine Learning Fundamentals: Concepts, Models, and Applications/_epub-src/OEBPS/xhtml/21_chapter08.xhtml) - 结论依据:摘录中关于类别编号不引入大小次序的定义、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:Next, we need to quantitively label the images with their corresponding outputs. In this case, we can label each image as the number that is written in the image. Thus, the label of a handwritten “0” would be 0, the label of a handwritten “1” would be a 1, and so on. However, we often use one-hot encoding to label the outputs for neural networks.