知识卡片
单权重梯度等于局部误差乘输入
若wᵏᵢⱼ从上一层i连到本层j,aᵏⱼ为激活前加权和,则∂E/∂wᵏᵢⱼ=δᵏⱼoᵏ⁻¹ᵢ,其中δ=∂E/∂a。先固定索引含义,可避免原文D.4把目标节点j误写i的混乱。
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
- · 附录D:反向传播
原书出处 · 1 处
- · 查看原书依据
Thus, the partial derivative of the error function E with respect to weight is We can see that the error E with respect to each individual weight in a layer k depends on the output of the layers coming before that. This is intuitive, as information propagates through the network. We will now also show that the error term actually depends on the error at
参考来源
- 位置:《Introduction to Autonomous Robots Mechanisms, Sensors, Actuators, and Algorithms (Nikolaus Correll, Bradley Hayes etc.)》附录D《附录D:反向传播》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Introduction to Autonomous Robots Mechanisms, Sensors, Actuators, and Algorithms (Nikolaus Correll, Bradley Hayes etc.)/_epub-src/OEBPS/xhtml/appendixD.xhtml)
- 结论依据:摘录中关于单权重梯度等于局部误差乘输入的定义、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:Thus, the partial derivative of the error function E with respect to weight is We can see that the error E with respect to each individual weight in a layer k depends on the output of the layers coming before that. This is intuitive, as information propagates through the network. We will now also show that the error term actually depends on the error at