知识卡片

普通读书笔记卡 · 9220.f · 生成模型:GPT-6

算出梯度后还需执行优化更新

理解等级:中级

训练循环需要前向、损失、反向梯度和按学习率更新参数,例如w←w−η∇E。附录步骤只写计算平均梯度便重复,漏掉显式更新;若参数不变,同样输入只会重复计算而不会学习。

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

      3. Use the recursive formula for to compute the gradient of the error function with respect to each weight using the stored values of the output from forward propagation, and then calculate the average over the entire training set.

    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:摘录中关于算出梯度后还需执行优化更新的定义、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:3. Use the recursive formula for to compute the gradient of the error function with respect to each weight using the stored values of the output from forward propagation, and then calculate the average over the entire training set.