知识卡片
没有非线性可合并卷积
在边界与尺寸兼容、忽略中间非线性的条件下,多个卷积级联可合成为一个较大支撑的线性卷积核。逐点非线性让中间组合依赖输入,增加表达能力,却仍可维持位置间参数共享。含偏置时要同时处理仿射项,有限边界处也未必等于单个统一核。
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
- · 卷积神经网络
原书出处 · 1 处
- · 查看原书依据
To understand this property, first imagine the CNN has no pointwise nonlinearities. Then the entire CNN is just the composition of a sequence of convolutions, which itself is a convolution (by equation (15.16), convolving a signal with multiple filters in a row is equivalent to convolving the signal with a single equivalent filter). Therefore, a CNN with no nonlinearities is itself just a single big
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:摘录中关于没有非线性可合并卷积的定义、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:To understand this property, first imagine the CNN has no pointwise nonlinearities. Then the entire CNN is just the composition of a sequence of convolutions, which itself is a convolution (by equation (15.16), convolving a signal with multiple filters in a row is equivalent to convolving the signal with a single equivalent filter). Therefore, a CNN with no nonlinearities is itself just a single big