← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 9168.c · 生成模型:GPT-6

偏置可写成恒定输入的权重

理解等级:中级

在输入向量前加x0=1,可把偏置b吸收入w0,统一写加权和。该技巧减少记号而不减少模型参数,更新时也要为恒定输入对应的偏置保留一致规则。

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

      Here, is the dot product, and the nonlinear activation function f(x) is also known as Heaviside step function. In practice, we are appending the value of “1” to the vector x so that x0 = 1, which simplifies wx + b (with w = {w1,…,wm}) to wx with w = {w0,w1,…,wm} where w0 takes the role of b. This is illustrated in figure 10.2, where

    参考来源

    - 位置:《Introduction to Autonomous Robots Mechanisms, Sensors, Actuators, and Algorithms (Nikolaus Correll, Bradley Hayes etc.)》第10章《人工神经网络》(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/chapter_10.xhtml) - 结论依据:摘录中关于偏置可写成恒定输入的权重的定义、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:Here, is the dot product, and the nonlinear activation function f(x) is also known as Heaviside step function. In practice, we are appending the value of “1” to the vector x so that x0 = 1, which simplifies wx + b (with w = {w1,…,wm}) to wx with w = {w0,w1,…,wm} where w0 takes the role of b. This is illustrated in figure 10.2, where