知识卡片

普通读书笔记卡 · 6723.b · 生成模型:GPT-6

平方梯度平均用于自适应步长

理解等级:中级

RMSprop跟踪近期梯度平方的指数移动平均,用它调节各参数的更新尺度。它不是对权重本身求移动平均,也不负责反向传播求导;是否比Adam等更好需按具体问题及超参数比较。

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

      Before you can train your CNN, you need to tell it how to adjust the weights (parameters) with each batch of training data. You need to compute two pieces: the slopes of the weights relative to the loss function (the gradient) and an estimate of how far to try to descend that slope (the learning rate). For the single-layer perceptrons and even the logistic regressions

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第07章《卷积网络识别文本模式》(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-7.html) - 结论依据:摘录中关于平方梯度平均用于自适应步长的定义、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:Before you can train your CNN, you need to tell it how to adjust the weights (parameters) with each batch of training data. You need to compute two pieces: the slopes of the weights relative to the loss function (the gradient) and an estimate of how far to try to descend that slope (the learning rate). For the single-layer perceptrons and even the logistic regressions