← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8863 · 生成模型:GPT-6

参数与激活承担不同职责

理解等级:初级

激活是某次输入经过网络得到的中间表示,参数是层所用的权重与偏置。二者虽然都可用张量表示,却不能仅由数据类型区分。常规训练调整参数,常规推理改变输入激活;分析网络时应标出哪些量固定、哪些量被优化。

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

      The activations are the neural representations of the data being processed. Often, we will not worry about distinguishing between inputs, hidden units, and outputs to the net, and simply refer to all data and neural activations in a network, layer by layer, as a sequence [x0, …, xL], in which case x0 is the raw input data.

    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:摘录中关于参数与激活承担不同职责的定义、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:The activations are the neural representations of the data being processed. Often, we will not worry about distinguishing between inputs, hidden units, and outputs to the net, and simply refer to all data and neural activations in a network, layer by layer, as a sequence [x0, …, xL], in which case x0 is the raw input data.