知识卡片
仿射层与非线性改变几何不同
仿射层按权重与偏置平移、旋转、伸缩或压缩点集;ReLU将负坐标压到零;L2归一化将非零向量投到单位球面。不同操作可能丢失不同信息,不能将所有层都视为可逆重排,或认为变换后的距离保持原样。
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
原书出处 · 1 处
- · 查看原书依据
Figure 13.5: 2D mapping diagrams for several neural layers. The linear layer mapping will shift, stretch, and rotate depending on its weights and biases. One interesting thing to notice here is that the relu layer maps many points to the axes of the positive quadrant. In general, with relu-nets, a lot of data density will build up along these axes, because any point not in
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第13章《作为分布变换器的神经网络》(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/c013.xhtml)
- 结论依据:摘录中关于仿射层与非线性改变几何不同的定义、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:Figure 13.5: 2D mapping diagrams for several neural layers. The linear layer mapping will shift, stretch, and rotate depending on its weights and biases. One interesting thing to notice here is that the relu layer maps many points to the axes of the positive quadrant. In general, with relu-nets, a lot of data density will build up along these axes, because any point not in