知识卡片

普通读书笔记卡 · 8932.b · 生成模型:GPT-6

调制可暴露频率捷径

理解等级:中级

把原有水平线乘空间余弦,会移动其频谱,使能量进入另一方向核偏好的频带,而人仍可能看见水平结构。书中玩具分类器因此产生错误,揭示训练分布允许的频率捷径。这个例子针对该模型,不是所有CNN都必然使用同一个捷径。

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

      Does the network solve the task that we had in mind? Can we predict which inputs will make the output fail? Can we produce test examples that to us look right but for which the network produces the wrong classification output? The goal of this analysis is to identify weaknesses in the learned representation and in our training set (missing training examples, biases in our

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第24章《卷积神经网络》(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/c024b.xhtml) - 结论依据:摘录中关于调制可暴露频率捷径的定义、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:Does the network solve the task that we had in mind? Can we predict which inputs will make the output fail? Can we produce test examples that to us look right but for which the network produces the wrong classification output? The goal of this analysis is to identify weaknesses in the learned representation and in our training set (missing training examples, biases in our