知识卡片
学习率控制沿梯度移动的尺度
基本下降更新沿负梯度,学习率控制步长,过大可能越过低损失区域,过小则进展缓慢。梯度的尺度还受输入、损失归约和样本权重影响;同一个学习率不保证不同配置有相同行为。
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
- · 神经网络基础
原书出处 · 1 处
- · 查看原书依据
Your optimizer needs two parameters to know how to or how quickly to ski along the loss slope: learning rate and momentum. The learning rate determines how much your weights change in response to an error—think of it as your “ski velocity.” Increasing it can help your model converge to the local minimum faster, but if it’s too large, you may overshoot the minimum every
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第05章《神经网络基础》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-5.html)
- 结论依据:摘录中关于学习率控制沿梯度移动的尺度的定义、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:Your optimizer needs two parameters to know how to or how quickly to ski along the loss slope: learning rate and momentum. The learning rate determines how much your weights change in response to an error—think of it as your “ski velocity.” Increasing it can help your model converge to the local minimum faster, but if it’s too large, you may overshoot the minimum every