知识卡片
宽度与深度改变表达效率
宽度增加每层可并行构造的特征,深度增加复合变换次数。某些函数族存在深度分离:较深网络可用远少于浅层网络的单元表达。但这是对特定函数族和架构的结论,不能推出所有任务越深越省参数或越易训练。
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
- · 神经网络
原书出处 · 1 处
- · 查看原书依据
there are certain classes of function that can be represented with a polynomial number of neurons in a depth d network but require an exponential number of neurons in a depth d′ network, for certain d′ < d. Arguments along these lines are called depth separations, and the interested reader can refer to [465] to learn more about this ongoing line of research.
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第12章《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/c012.xhtml)
- 结论依据:摘录中关于宽度与深度改变表达效率的定义、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:there are certain classes of function that can be represented with a polynomial number of neurons in a depth d network but require an exponential number of neurons in a depth d′ network, for certain d′ < d. Arguments along these lines are called depth separations, and the interested reader can refer to [465] to learn more about this ongoing line of research.