知识卡片

普通读书笔记卡 · 9001.b · 生成模型:GPT-6

舌头共现变成捷径

理解等级:中级

书中上色网络常在狗嘴下面添加粉红色:训练集里狗与伸出的舌头经常同时出现,网络学到了易用的统计捷径。检查错误时要追查共现背景、姿态和局部特征;类别出现得多,也不等于该类的关键变化覆盖充分。

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

      It adds a pinkish blob underneath their mouths! Why? Because not only is ImageNet full of dogs, it is full of dogs with their tongues out. So the net learned that it's a good strategy to put down a blob of pink whenever it detects a dog mouth, regardless of whether that mouth is open or closed.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第35章《数据偏差与分布漂移》(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/c035b.xhtml) - 结论依据:摘录中关于舌头共现变成捷径的定义、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:It adds a pinkish blob underneath their mouths! Why? Because not only is ImageNet full of dogs, it is full of dogs with their tongues out. So the net learned that it's a good strategy to put down a blob of pink whenever it detects a dog mouth, regardless of whether that mouth is open or closed.